Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: sgwrx on April 05, 2006, 04:37:18 pm
-
http://www.rags-int-inc.com/PhotoTechStuff/ (http://www.rags-int-inc.com/PhotoTechStuff/)
the one i'm referring to is called "Tones n Zones"
-
Some interesting information in the article. I didn't get a chance to give it a full read so pardon me if I missed something. If I have time latter, I'll read it in full and respond more if needed.
The most serious misconception is that because of the computer’s binary ordering...
He is misinterpreting (or reacting to someone who misinterpreted) the purpose of that chart. It is not representing anything that results from binary ordering. The chart, which I recognize from Michael's Expose to the Right and Raw file articles (I've also used it in my Bit-Depth article), describes the linear nature of how camera sensors capture data. His discussion seems to evolve around a gamma 2.2 encoded file which I didn't see anything to object to.
-
Quote from the article:
"Over exposure will clip tonal details quickly. Under exposure also distorts tonal values, but it retains shadow detail longer and can be recovered easier. This should not be news to any digital photographer. "
So he's proposing an "expose to the left" strategy?
-
I'm marking this for reading later, but if indeed he is proposing an expose-to-the-left priority of exposure, then I'm going to have to respectfully disagree based on my practical experience. Exposing to the left and then pulling up in RAW conversion only increases noise and posterization, while exposing normally (or better yet in most cases, to the right) does not make these issues more acute.
-
He's not proposing expose to the left, but expose for midtones.
-
It's an interesting, but difficult to follow article, where the conclusion doesn't seem to quite follow from the body of the argument. Gamma is always an interesting subject as it usually performs the dual purpose of perceptually encoding a higher dynamic range into a lower number of bits (ie 12bit sensor to 8bit data) and being the inverse characteristic to the traditional CRT monitor.
Graeme
-
i've re-read "expose to the right" and the article i've posted, i'm still working this all out.
but one question, "expose to the right" seems to be taken within context of signal to noise ratio. would the following be an accurate application / example of "expose to the right":
one takes a photograph of a black leather couch against a medium gray wall.
if one exposes so that the black couch appears black, the couch would contain noise.
if one exposed to the right, making the black couch appear more towards white or middle gray, the couch would contain much less noise.
finally, in post processing one would adjust the raw image so that the shifted-towards-white/gray couch appeared more black.
the resulting image would contain much less noise than exposing the couch according to an in-camera meter's suggestion.
-
i've re-read "expose to the right" and the article i've posted, i'm still working this all out.
but one question, "expose to the right" seems to be taken within context of signal to noise ratio. would the following be an accurate application / example of "expose to the right":
one takes a photograph of a black leather couch against a medium gray wall.
if one exposes so that the black couch appears black, the couch would contain noise.
if one exposed to the right, making the black couch appear more towards white or middle gray, the couch would contain much less noise.
finally, in post processing one would adjust the raw image so that the shifted-towards-white/gray couch appeared more black.
the resulting image would contain much less noise than exposing the couch according to an in-camera meter's suggestion.
[a href=\"index.php?act=findpost&pid=62039\"][{POST_SNAPBACK}][/a]
I think you've summed it up well.
Expose to the right isn't about over exposing (duh), it's about exposing for a linear encoded file. That's kind of unique with RAW (everything else is gamma corrected). So ensuring you're not blowing out highlights and placing as much good data within that linear scale is what you want to try to accomplish. Film (the old stuff we're used to) wasn't linear encoded (remember those good old H&D curves?).
-
I think you've summed it up well.
Expose to the right isn't about over exposing (duh), it's about exposing for a linear encoded file. That's kind of unique with RAW (everything else is gamma corrected). So ensuring you're not blowing out highlights and placing as much good data within that linear scale is what you want to try to accomplish. Film (the old stuff we're used to) wasn't linear encoded (remember those good old H&D curves?).
[a href=\"index.php?act=findpost&pid=62054\"][{POST_SNAPBACK}][/a]
For a digital camera that only supports jpeg and not RAW, can you clarify the following:
- can this "expose to the right" technigue still be used?
- can the RAW conversion sw and methods (e.g. merging to expand dynamic range, etc.) be used on jpegs?
Thanks.
-
For a digital camera that only supports jpeg and not RAW, can you clarify the following:
- can this "expose to the right" technigue still be used?
- can the RAW conversion sw and methods (e.g. merging to expand dynamic range, etc.) be used on jpegs?
Thanks.
[a href=\"index.php?act=findpost&pid=62272\"][{POST_SNAPBACK}][/a]
No, the expose to the right is specifically for linear encoded data (RAW). A JPEG is an existing rendered 8-bit file. You get what you get. That's the beauty of RAW.
-
For a digital camera that only supports jpeg and not RAW, can you clarify the following:
- can this "expose to the right" technigue still be used?
- can the RAW conversion sw and methods (e.g. merging to expand dynamic range, etc.) be used on jpegs?
Thanks.
[a href=\"index.php?act=findpost&pid=62272\"][{POST_SNAPBACK}][/a]
"Expose to the right" can still be used; I was doing it this weekend in sand-dunes. Basically, if you check the histogram and a substantial part at the right is flat zero, and a longer exposure time is acceptable, then you try again with increased exposure (+ compensation?) to improve shadow handling. You might then need to do some post-processing to bring tones down.
I believe that merging multiple frames to expand dynamic range is also possible; there is no fundamental reason why that would not be possible with JPEG.
-
I believe that merging multiple frames to expand dynamic range is also possible; there is no fundamental reason why that would not be possible with JPEG.
[a href=\"index.php?act=findpost&pid=62296\"][{POST_SNAPBACK}][/a]
There is a great chunk of high dynamic range software that only works with JPEG files - I think even CS2 does some conversion on the RAW files, whether that is to JPEG or not I can't remember. Prior to merging successive frames the software needs to convert the JPEG data gamma corrected and post application of tone curve to linear data (a kind of 'unbending' the data). RAW, linear, JPEG, TIFF it is possible to use all types of files as input for high dynamic range images.
-
http://www.rags-int-inc.com/PhotoTechStuff/ (http://www.rags-int-inc.com/PhotoTechStuff/)
the one i'm referring to is called "Tones n Zones"
[a href=\"index.php?act=findpost&pid=61937\"][{POST_SNAPBACK}][/a]
Rags Gardner unfortunately likes to take an engineer's background and apply it to something he really doesn't understand-photographic exposure. He still thinks that digital should be "exposed for the mid-tones" ala film and the "old Zone System".
Let's see...on one hand we have Thomas Knoll, Bruce Fraser, Michael Reichmann. On the other side is Rags Gardner-who do you believe? I think I'll go with Thomas...in my experience he knows this stuff inside and out from the standpoint of digital sensors.
Rags writes: "So, let’s start with the fact that an image can be divided into lighting zones, as taught by Ansel Adams."
Well, that's where his whole argument falls down. You can't get to the zones until AFTER you apply a tone curve and that's where the distribution of tones from a linear capture gets turned into a gamma encoded image. Ever see a linear capture processed for linear gamma? Looks really dark and under exposed. It bares no relationship to a standard gamma encoded image whether digital or film.
Rags wrote this, uh, piece of uh, whatever in direct response to an online dispute with Bruce Fraser at the Adobe Camera Raw forums...it looks authoritative, but it's fundamentally flawed. Digital capture is recorded in linear manner and it's not until the linear image is demosaiced and tone curved that it represents an image one can see and use. Rags wants to try to convince people to meter and expose for zone V which worked with B&W film but for digital capture you should meter to just maintain textural highlight detail-much more like zone IX if anything.
-
It is correct to expose for highlight and ensure they don't clip, but don't waste bits either. You're right that linear gamma images look dark and underexposed, and that they need both a gamma correction and a s-curve to look correct.
Graeme
-
It is correct to expose for highlight and ensure they don't clip, but don't waste bits either. You're right that linear gamma images look dark and underexposed, and that they need both a gamma correction and a s-curve to look correct.
Graeme
[a href=\"index.php?act=findpost&pid=62501\"][{POST_SNAPBACK}][/a]
The don't look dark when you assign a profile for that capture. But the Histogram looks all pushed to one side due to the gamma encoding (which after a proper conversion to a working space levels out). I have linear encoded RGB files that look dark and ugly only up to the point I assign the proper profile, then they look just fine.
-
The don't look dark when you assign a profile for that capture. But the Histogram looks all pushed to one side due to the gamma encoding (which after a proper conversion to a working space levels out). I have linear encoded RGB files that look dark and ugly only up to the point I assign the proper profile, then they look just fine.
But what is that assigning of a profile doing, image processing-wise?
Graeme
-
But what is that assigning of a profile doing, image processing-wise?
Graeme
[a href=\"index.php?act=findpost&pid=62512\"][{POST_SNAPBACK}][/a]
It defines the scale of the numbers so Photoshop properly previews the numbers. That's all a profile does (define numbers within a scale of human vision).
-
Rags Gardner unfortunately likes to take an engineer's background and apply it to something he really doesn't understand-photographic exposure. He still thinks that digital should be "exposed for the mid-tones" ala film and the "old Zone System".
Rags writes: "So, let’s start with the fact that an image can be divided into lighting zones, as taught by Ansel Adams."
Well, that's where his whole argument falls down. You can't get to the zones until AFTER you apply a tone curve and that's where the distribution of tones from a linear capture gets turned into a gamma encoded image. Ever see a linear capture processed for linear gamma? Looks really dark and under exposed. It bares no relationship to a standard gamma encoded image whether digital or film.
If anything, the zones are easier to work with in linear since each zone is exactly twice the value of the preceeding. Each stop of extra exposure doubles the pixel value. The relationships are well demonstrated on Norman Koren's site:
http://www.normankoren.com/digital_tonality.html (http://www.normankoren.com/digital_tonality.html)
As Mr. Rodney noted, a gamma one image looks just fine in Photoshop if the proper profile is attached, but that is true for other spaces as well.
-
If anything, the zones are easier to work with in linear since each zone is exactly twice the value of the preceeding.
[a href=\"index.php?act=findpost&pid=62523\"][{POST_SNAPBACK}][/a]
If you want to work in a linear space and make a custom Photoshop working space with a gamma 1.0, then yes. But the big point of failure with Rag's 18% midtone approach is in linear space, a middle grey-the target Rag's says to shoot for is about level 50 in an 8 bit linear file. Way too low to do any kind of "normal" tone curve and it's pretty far south of optimal in terms of midtone. And if you examine the relative levels consentration of a gamma 1.0 image there is WAY too many levels packed into the brightest bits (the expose to the right concept) and a gamma 1.0 image in 8 bit/channel would be extremely prone to breaking.
The raw image capture is in linear and has far more bits (levels) to deal with in the highlights. The shadows far less bits. Therefore it makes far more sense to expose for the textural highlights and make use of all those bits. It's far easier and produces much better signal to noise to keep the midtone up the scale not down the scale. You can easily make things darker without adding noise than the other way around.
-
A big breakthrough in dynamic range is long overdue. The current linear capture system provides far too many levels for the brighter parts of the image and far too few levels for the darker tones. If the scene is of relatively low dynamic range, consisting mainly of midtones, there's no great need to expose to the right. Correct exposure for the mid tones will suffice. If the DR of the scene is wide, as many landscape shots are which include sky or sunlit areas, then exposing to the right is pretty essential for noise-free shadow detail.
Since I'm not an engineer, I find it difficult to appreciate the difficulties involved in devising a capture system which compresses the DR. That is, one which redistributes the levels through a process of selective augmentation of low level signals.
We already have an example of excellent noise reduction with Canon DSLRs at high ISO. Take two shots using the same exposure, but one at ISO 100 and the other at ISO 1600, then compare the shadows. The ISO 1600 shot will likely have much better shadow detail, yet those shadows (on the sensor) in both shots have received the same amount of light. It seems that amplification of the analog signal prior to digitisation allows for dramatic noise reduction. Of course, if the exposure at ISO 100 was a full exposure to the right, then the same exposure at ISO 1600 would blow the highlights by 4 stops of overexposure.
But supposing there was a way of diminishing the intensity of those 4 stops, ie. the darker tones are augmented and the brighter tones are simultaneously diminished. We would then have a compression of dynamic range which, when unpacked, could be very wide indeed.
I'm reminded of developments in vinyl LP audio recording prior to the audio CD. The dynamic range of LP discs used to be typically around 50dB until a system called dbx was invented which compressed the signal at the recording stage, uncompressed it during playback producing a significant boost to DR which, as I recall, was around 80dB. However, CD audio gave us around 90dB and greater, so the dbx technology was too late.
-
If the scene is of relatively low dynamic range, consisting mainly of midtones, there's no great need to expose to the right. Correct exposure for the mid tones will suffice. If the DR of the scene is wide, as many landscape shots are which include sky or sunlit areas, then exposing to the right is pretty essential for noise-free shadow detail.
[a href=\"index.php?act=findpost&pid=62531\"][{POST_SNAPBACK}][/a]
Actually, you have it backwards...in the case of a scene that DOES fit in the dynamic range of a sensor, you are GOING to get far smoother captures by increasing the exposure and placing more of the scene in the lighter tones-as long as you don't clip textural highlight detail.
In the case of a scene that exceedes the dynamic range of the sensor, your only choice is to decide, your self, what part of the scene is the most important and expose as best you can knowing the bright specular will clip. The ONLY way around this is to either shoot multiple exposures to assemble after the fact or do a dual process from raw to regain as much highlight detail as possible and blend the highlights into a normally processed image.
The other factor Rags seems to ignore is that sensors wiill indeed vary camera to camera-although a single camera will itself be pretty constistant.
When you set the ISO to 100, it's a nominal setting. Your actual sensor sensitivity might be 80, or 120. This can indeed impact the way your exposures will be biased when shooting. Using an exposure compensation will help nail exposures better based upon the REAL ISO of your camera, not the nominal ISO. Once to nail that, the next step is to learn the exact dynamic range of YOUR sensor. Is it 6, 7, 8 REAL stops? Depending on your acceptance of noise, one can make an argument that many DSLR are at or near 8 stops, maybe a bit more-particularly when you examine just how much data is there nclumped, not clipped at the highlights if you can get to it.
Camera Raw does a really remarkable extraction of highlight detail, why? Aside from the fact that Camera Raw DOESN'T quit when the first channel is clipped, the fact is there's an enourmous amount of detail in those extreme highlights. It's there if you know how to use it.
Which, if you approach exposure based upon metering for the mid-tones and letting highlights and shadows fall will pretty much make sure that you loose that advantage of controlling the usefulness of your dynamic range and your bits.
Back when sensors first came out, there was a serious problem called blooming that would have photosites in a sensor woth extreme spectrals bleed accross to other photosites and creally blow out. In that period a photographer HAD to "under expose" digital to keep from having speculars blow accross large portions of the sensor. Those times are effectively ended with today's sensors which simply do not bloom like the older ones did.
The problem really boils down to one of convention...digital sensors are NOT like film. They do not react the way chrome nor neg film reacts to light. Therefore you really can't use old film exposure techniques...the aim is to get as much data captured as far up the scale as you can while still preserving usable textural highlight detail. This is NOT "over exposing" it's "proper exposing"-for a digital sensor.
-
Actually, you have it backwards...in the case of a scene that DOES fit in the dynamic range of a sensor, you are GOING to get far smoother captures by increasing the exposure and placing more of the scene in the lighter tones-as long as you don't clip textural highlight detail.
[a href=\"index.php?act=findpost&pid=62533\"][{POST_SNAPBACK}][/a]
I was referring to a scene well within the DR of the sensor, not one which fits it. In other words, if the histogram does not occupy regions close to the left or right, but is largely in the middle, there's not much point in increasing exposure to push it more to the right. Doing so will give you more levels, but you probably already have enough. Exposing to the right provides sorely needed levels on the left.
-
If you want to work in a linear space and make a custom Photoshop working space with a gamma 1.0, then yes. But the big point of failure with Rag's 18% midtone approach is in linear space, a middle grey-the target Rag's says to shoot for is about level 50 in an 8 bit linear file. Way too low to do any kind of "normal" tone curve and it's pretty far south of optimal in terms of midtone. And if you examine the relative levels consentration of a gamma 1.0 image there is WAY too many levels packed into the brightest bits (the expose to the right concept) and a gamma 1.0 image in 8 bit/channel would be extremely prone to breaking.
The raw image capture is in linear and has far more bits (levels) to deal with in the highlights. The shadows far less bits. Therefore it makes far more sense to expose for the textural highlights and make use of all those bits. It's far easier and produces much better signal to noise to keep the midtone up the scale not down the scale. You can easily make things darker without adding noise than the other way around.
[a href=\"index.php?act=findpost&pid=62526\"][{POST_SNAPBACK}][/a]
I agree with Jeff's analysis, but my point was that one can use a zone system approach with digital, and it makes a lot of sense. With his negative material, Adams exposed for the shadows, not the midtones. With digital, I would expose for Zone 1.
Converting the raw to gamma one, sometimes with no white balance, with DCRaw is often instructive, but in practical work it is easier to work in a gamma encoded space.
-
We already have an example of excellent noise reduction with Canon DSLRs at high ISO. Take two shots using the same exposure, but one at ISO 100 and the other at ISO 1600, then compare the shadows. The ISO 1600 shot will likely have much better shadow detail, yet those shadows (on the sensor) in both shots have received the same amount of light. It seems that amplification of the analog signal prior to digitisation allows for dramatic noise reduction. Of course, if the exposure at ISO 100 was a full exposure to the right, then the same exposure at ISO 1600 would blow the highlights by 4 stops of overexposure.
Ray, the ISO 1600 exposure will have markedly less dynamic range and much more noise than the ISO 100 image because with the short exposure in the ISO 1600 image, few photons were collected by the sensor, and the number of photons converted to electrons is the main determinant of noise and DR. For a scientific analysis of DR and noise, see this post and look at table 1:
http://www.clarkvision.com/imagedetail/eva...-1d2/index.html (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html)
The shadows and all other levels in the high ISO exposure do not receive the same amount of light as with the low ISO exposure--the high ISO exposure will be less for all levels.
-
I was referring to a scene well within the DR of the sensor, not one which fits it. In other words, if the histogram does not occupy regions close to the left or right, but is largely in the middle, there's not much point in increasing exposure to push it more to the right. Doing so will give you more levels, but you probably already have enough. Exposing to the right provides sorely needed levels on the left.
[a href=\"index.php?act=findpost&pid=62534\"][{POST_SNAPBACK}][/a]
Ray, I agree that in the exposure situation to which you refer, there may already be enough levels to represent the scene. Twelve bit raw capture can represent 4096 levels and 256 levels are sufficient for most images. If you place your highlight at 2048 (down one stop), you still have 2048 levels to work with. I think that proponents of ETTR overplay the importance of levels.
However, if you place your highlight at 2048 rather than 4095 (with a Canon EOS 1D Mark II) your signal to noise drops from 256 to 172 according to Roger Clark's calculations. The dynamic range (maximum signal / read noise) suffers a similar hit. IMHO, signal to noise and dynamic range are often a better argument for ETTR than the number of levels. The same considerations apply down the tonal scale to from the lighlights to the shadows, and the effects are most pronounced in the shadows.
-
Since I'm not an engineer, I find it difficult to appreciate the difficulties involved in devising a capture system which compresses the DR. That is, one which redistributes the levels through a process of selective augmentation of low level signals.
But supposing there was a way of diminishing the intensity of those 4 stops, ie. the darker tones are augmented and the brighter tones are simultaneously diminished. We would then have a compression of dynamic range which, when unpacked, could be very wide indeed.
I'm reminded of developments in vinyl LP audio recording prior to the audio CD. The dynamic range of LP discs used to be typically around 50dB until a system called dbx was invented which compressed the signal at the recording stage, uncompressed it during playback producing a significant boost to DR which, as I recall, was around 80dB. However, CD audio gave us around 90dB and greater, so the dbx technology was too late.
[a href=\"index.php?act=findpost&pid=62531\"][{POST_SNAPBACK}][/a]
Ray,
Your posts have raised a lot of good questions and have inspired me to reread some sources and do some thinking about the issues. In fact, analog to digital converters that use a log scale rather than linear are available and are widely used in signal processing--they compress the high signal levels rather than augmenting the lower levels, but the effect is the same. These methods are used when bandwidth is a major consideration, as in telephony.
However, such log AD converters are not used in any digital raw capture that I know of. One can either compress the image (such as is done with a gamma correction in 8 bit JPEGs) or use a sufficient number of bits to represent the entire dynamic range with a sufficient number of levels in the shadows as is done with raw capture. Most cameras use 12 bits, but some high end models use 16. As dynamic range increases, 16 bit encoding will probably become more common.
As Bruce Fraser explains in some depth in his Camera Raw with PSCS2 book, many operations such as white balance, highlight recovery, etc, are best done on the linear image data. The reason that ETTR applies mainly to raw capture is that one can simply use a linear exposure adjustment in ACR to lower the whole image tone values by a set amount, say one stop, in order to correct the mid tones to where they should be in a short scale subject exposed to the right. If a gamma correction had been applied, you would have to use some type of curve to adjust the tones nonlinearly. Also, with 8 bit encoding you have thrown away so many tone values that the adjustment could not recover the original data.
With regard to your sound recording analogy, the current approach in recording is simply to use more bits rather than compression. When storage becomes a consideration, as with iPods, compression is used.
-
The difference with audio though, is that we can see a much greater dynamic range that we can hear. CD's with their 96db dynamic range are pure overkill. Considering it's very hard to get background noise in a typical home lower than what, 40 to 50db, and that going much beyond 100db is painful, the 55db or so of vinyl is more than adequate.
Graeme
-
The difference with audio though, is that we can see a much greater dynamic range that we can hear. CD's with their 96db dynamic range are pure overkill. Considering it's very hard to get background noise in a typical home lower than what, 40 to 50db, and that going much beyond 100db is painful, the 55db or so of vinyl is more than adequate.
Graeme
[a href=\"index.php?act=findpost&pid=62565\"][{POST_SNAPBACK}][/a]
Quite true. The situation is even worse when listening to a CD in the car. With classical music, the softer passages are inaudible over the wind noise. My old car had dynamic range compression, which helped, but unfortunately my new one does not.
With photographs, however, most would prefer more dynamic range even at the expense of megapixels.
-
In other words, if the histogram does not occupy regions close to the left or right, but is largely in the middle, there's not much point in increasing exposure to push it more to the right. Doing so will give you more levels, but you probably already have enough.
[a href=\"index.php?act=findpost&pid=62534\"][{POST_SNAPBACK}][/a]
Well, that's where I think you are wrong...increasing the exposure to get the main portion of the histogram to the right -WILL- provide more tones (levels) and allow you to do more with the data such as expand the data and increase contrast on the resulting raw conversion WITHOUT the noise growing in the shadows.
As long as you don't clip textural highlights you -WILL- get better results (better signal to noise) by increasing the exposure. Since there are more levels the further up the scale you go, you'll have more levels to control and use. The curves can get pretty tweaky at times and you may need to do a dual or multi process conversion but the more levels you have in the final file the smoother the resulting tone curved image will be.
Under expose and you increase the noise, increase the exposure and you reduce the noise...the final toned file may look real close but the increased exposure will provide a cleaner file. Assuming your increased exposure didn't produce camera shake or reduced depth of field resulting in other non-digital image defects.
-
I'm with Schewe here. When shooting a scene within the DR of the sensor, You'll always get a better result by pushing exposure to the right as long as you don't do so to the point of clipping. The number of levels involved may be overkill regardless of how you expose, but the signal-to-noise ratio will be greater when exposure is pushed to the right. The difference may be small in many cases, but there will always be a difference.
-
When an image fits within the camera's DR, that is the best opportunity to utilize ETTR to its greatest potential.
-
And when it does not, you still must use your knowledge of the behavior of the sensor to optimize what you capture vs what gets lost. ETTR is the best paradigm for figuring that out when using a digital sensor. The zone system worked well for B&W film, but is unsuitable for digital because of the difference between the TRC's of film and digital sensors.
-
The current linear capture system provides far too many levels for the brighter parts of the image and far too few levels for the darker tones.
[a href=\"index.php?act=findpost&pid=62531\"][{POST_SNAPBACK}][/a]
Linearity really only enters at A/D conversion: until then you have just RAW data in the form of photon counts. And it seems that there is little problem producing an A/D converter (say 16 bits) than can handle the full DR of any sensor's output with no significant loss of sensor information, linearity is not a problem at that stage. Nor is there much penalty to storing the few extra bits related to excessively fine tonal distinctions in bright parts of the image.
So the most useful place to introduce nonlinearity is in the subsequent encoding into formats with bit depth less than the number of stops of DR --- which is what gamma compression and tone curves do.
P. S. (not to Ray in particular) It is a myth that gamma compression is needed to match the human visual system: it is needed to compensate for the fact that monitors perform gamma expansion, with a very nonlinear relationship between input voltage and output brightness. Native CRT gamma is about 2.5, with calibration bringing it down to 2.2 in the Windows standard and 1.8 in the Mac standard. (Printers also have a native gamma, which is typically about 1.8, or at least was when Apple chose their gamma of 1.8.)
So if you have files converted with linear gamma, setting your monitor calibration to linear gamma will display them correctly. I create a linear gamma profile option on my Macs in System Preferences>Displays>Color for purposes like this.
-
When an image fits within the camera's DR, that is the best opportunity to utilize ETTR to its greatest potential.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=62636\")
And when it does not, you still must use your knowledge of the behavior of the sensor to optimize what you capture vs what gets lost. ETTR is the best paradigm for figuring that out when using a digital sensor. The zone system worked well for B&W film, but is unsuitable for digital because of the difference between the TRC's of film and digital sensors.
[a href=\"index.php?act=findpost&pid=62640\"][{POST_SNAPBACK}][/a]
I find Jonathan's comments confusing. ETTR applies only when the dynamic range of that portion of the scene you wish to capture is less than that of the sensor. This is what Ansel Adams calls a short scale subject. With negative film, Adams exposed to the left, but with digital, we expose to the right. In both cases, we are making best use of the properties of the medium. If the dynamic range of the scene equals that of the sensor, there is no room for ETTR. If the dynamic range of the subject exceeds that of the sensor, as Jeff Schewe points out above, you have to expose for the most important part of the scene and accept clipping or else use multiple exposures and combine them with HDR in Photoshop or by other means. ETTR does not remap the values or tell us how out of range values should be handled.
The digital sensor is indeed linear, wheras film's response is log. However, when we apply a gamma correction and a TRC to the digital, the response is not unlike that of film. Zone concepts can easily be applied to digital capture, as discussed by Norman Koren in the links below. He even gives equations to convert pixel values to zones.
[a href=\"http://www.normankoren.com/digital_tonality.html#Gamma_contrast]http://www.normankoren.com/digital_tonalit...#Gamma_contrast[/url]
http://www.normankoren.com/zonesystem.html (http://www.normankoren.com/zonesystem.html)
Furthermore, if you look at the characteristic curve of a digital image processed in camera or in Photoshop with Norman's Imitest, it looks very much like the H&D curves of film. Concepts we learned with the zone system serve us very well today, but are much easier to apply with digital.
-
Concepts we learned with the zone system serve us very well today, but are much easier to apply with digital.
[a href=\"index.php?act=findpost&pid=62666\"][{POST_SNAPBACK}][/a]
While the concepts may serve us, the exact implementation is -NOT- the same and that's where some people, such as Rags goes astray.
Rags is still trying to force fit analog film techniques to digital capture and that dog don't hunt.
Likewise, the Zone System pretty much falls apart because when dealing with linear captures there is an unprecedented amount of detail in the brightest potion of a linear capture. To the point that while it may SEEM like data may be getting clipped, use of a negative Exposure setting reveals far more data to be had via Camera Raw than any experience with film would indicate.
The Zone system approach really only works post raw processing in working with tone reproduction curves and even then it's not really a direct application since the old days of fixed contrast, or even poly-contrast silver gel papers no longer applies.
The Zone system as I learned it (at RIT under Zakia, Todd and Strobel) was a method of mapping scene contrast range to negative contrast to paper contrast in a manner that allowed one to pre-visualize the relative B&W reproduction of the original scene. That said, Ansel Adams had no problem using altered neg development, local hot paper developing, dodging and burning and anything else he had to do in the darkroom to get the exact tone reproduction he desired.
Knowledge of the zone system may be useful but I would be careful assuming there is a direct relationship to digital capture. Photoshop and even Camera Raw completely alters the old technical limitations of B&W neg film, developers and traditional silver paper and developers.
Again, I must point out the importance of determining your sensor's exact ISO and dynamic range and to expose to just retain textural detail without clipping. But without certain knowledge of your sensor's ISO and dynamic range, most people fail to actually use that densely packed area of near clipped data.
-
While the concepts may serve us, the exact implimentation is -NOT- the same and that's where some peope, such as Rags goes astray.
The Zone system as I learned it (at RIT under Zakia, Todd and Strobel) was a method of mapping scene contrast range to negative contrast to paper contrast in a manner that allowed one to pre-visualize the relative B&W reproduction of the original scene. That said, Ansel Adams had no problem using altered neg development, local hot paper developing, dodging and burning and anything else he had to do in the darkroom to get the exact tone reproduction he desired.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=62676\")
Of course, I would agree that the implementation is quite different. One should also note that the zone system is for monochrome photography. Although this thread started with reference to Rags' exposition on his web site, I don't think that it is necessary to rehash certain technical errors in Rags' arguments--these have already been discussed on the Adobe Camera Raw Forum (Exposure to the Right and Tone Placement). Why this fixation on Rags?
With digital, we are still faced with mapping scene contrast to tonal values in a print (assuming the print is the final product, which in many cases is no longer true). As explained in a white paper on the Adobe web site (which uses one of your pictures for illustration, by the way), even the best print has a contrast ratio of 500:1, whereas the original scene may have a contrast ratio of 100,000:1. As the author points out, the art of photography is the interpretation of the scene on the printed page. A literal representation is not possible.
[a href=\"http://www.adobe.com/digitalimag/pdfs/calibrating_digital_darkroom.pdf]http://www.adobe.com/digitalimag/pdfs/cali...al_darkroom.pdf[/url]
In this re-mapping, it can be useful to divide the scene into 10 or so zones similar to what Adams used in the zone system and then deal with the zones digitally. One Photoshop plugin (which I have not used) goes so far as to place the zones (roman numerals and all) on its interface:
http://www.curvemeister.com/ (http://www.curvemeister.com/)
As you rightfully stress, the process should begin with calibration of the light meter. In his The Negative, Adams preceeds the zone discussion with calibration of the meter. Some concepts still apply. However, for digital exposure should be based on Zone X, not Zone IV. With highlight recovery in ACR, some highlight clipping can be tolerated, and one can expose even more to the right for more dynamic range.
However, as Norman Koren points out: "In a photographic print, which has about a 100:1 luminance ratio, the eye can distinguish between 100 and 200 discrete luminance levels-- fewer than the 256 available in 8-bit B&W or 24-bit color". In most cases one can not really make full use of all those 2048 tones in the brightest stop of a 12 bit digital capture .
http://www.normankoren.com/digital_tonality.html (http://www.normankoren.com/digital_tonality.html)
-
As you rightfully stress, the process should begin with calibration of the light meter. In his The Negative, Adams preceeds the zone discussion with calibration of the meter.
Well, I would argue that the calibration is not just the meter. . .but the entire exposure system-particularly the sensor and it's actual vs. nominal sensitivity and the exact dynamic range it's capable of. Which is different than Adam's calibration of just the meter...but if you are talking about the concept of gaining control over the entire "system", then I agree.
In most cases one can not really make full use of all those 2048 tones in the brightest stop of a 12 bit digital capture .
No, you can't make use of -ALL- the levels of the brightest stop of detail but with proper toning you -CAN- use a lot more than most people think they can. I would point people to an article on the Adobe site that talks about Highlight Recovery in Camera Raw (http://www.adobe.com/digitalimag/pdfs/highlight_recovery.pdf) (2.7MB PDF) (Which will be updated for Camera Raw 3 as soon as Adobe gets around to posting the revised article-it's been done for a couple of months now)
Why this fixation on Rags?
It's not so much that I have a fixation on Rags as the article in question is the basis of this thread so I like to try to stay on topic where I can...that and the fact that Rags has gone to the lengths he has and I'm afraid too many people will drink his Cool-Aid, ya know?
:~)
-
However, as Norman Koren points out: "In a photographic print, which has about a 100:1 luminance ratio, the eye can distinguish between 100 and 200 discrete luminance levels-- fewer than the 256 available in 8-bit B&W or 24-bit color". In most cases one can not really make full use of all those 2048 tones in the brightest stop of a 12 bit digital capture.
http://www.normankoren.com/digital_tonality.html (http://www.normankoren.com/digital_tonality.html)
[a href=\"index.php?act=findpost&pid=62699\"][{POST_SNAPBACK}][/a]
Mr. Koren isn't fully accurate in that. While true, the human eye can distinguish 100 to 200 discrete tones, they are also distinct tones. The difference between 250 and 249, for example, is so small the eye won't notice it. There needs to be a good difference of about three tonal values in a 24-bit image before you'll readily notice a difference. It would be more accurate to say they eye can distinguish only 5/12th (108 out of 256) to 1/3rd (83 out of 256) of the 256 available tones in a 24-bit image. That puts a 24-bit at the least in the low end of the 100-200 the eye can distinguish.
-
Well, I would argue that the calibration is not just the meter. . .but the entire exposure system-particularly the sensor and it's actual vs. nominal sensitivity and the exact dynamic range it's capable of. Which is different than Adam's calibration of just the meter...but if you are talking about the concept of gaining control over the entire "system", then I agree.
[a href=\"index.php?act=findpost&pid=62715\"][{POST_SNAPBACK}][/a]
Yes, I should have specified exposure system calibration, not merely meter calibration. In practice, the meter has to be reproducable but not necessarily well calibrated--one merely needs to know the exposure bias from the metered reading needed to give the maximum data number in the raw file (usually about 4095, but sometimes less) and use this to place the highlight.
Knowledge of the dynamic range is helpful so you will have some idea of the shadow tonal values and noise, but the effective dynamic range varies markedly with the ISO setting and noise characteristics of the camera. A lot of testing is needed to get these data so many of us simply use the lowest practical ISO and hope for the best.
-
Ray, the ISO 1600 exposure will have markedly less dynamic range and much more noise than the ISO 100 image because with the short exposure in the ISO 1600 image, few photons were collected by the sensor, and the number of photons converted to electrons is the main determinant of noise and DR. For a scientific analysis of DR and noise, see this post and look at table 1:
The shadows and all other levels in the high ISO exposure do not receive the same amount of light as with the low ISO exposure--the high ISO exposure will be less for all levels.
[a href=\"index.php?act=findpost&pid=62545\"][{POST_SNAPBACK}][/a]
I'm still travelling so don't often get on the net. I see you've misunderstood what I mean here. Exposure is determined by shutter speed and f stop. No matter what the ISO, if the exposure is the same, then the same amount of light falls upon the sensor. An ISO setting is merely an instruction to the camera to amplify the signal and apply certain noise reduction processes. If you compare a 'correctly' exposed shot at ISO 1600 (exposed fully to the right) with the same shot at ISO 100 (same shutter speed and f stop), the ISO 100 shot will appear to be underexposed. However, the sensor has received the same amount of light, yet shadow noise will be very much greater. This implies that noise reduction cna be more successfully implemented when the signal is boosted.
-
Well, that's where I think you are wrong...increasing the exposure to get the main portion of the histogram to the right -WILL- provide more tones (levels) and allow you to do more with the data such as expand the data and increase contrast on the resulting raw conversion WITHOUT the noise growing in the shadows.
[a href=\"index.php?act=findpost&pid=62575\"][{POST_SNAPBACK}][/a]
Jeff,
I don't think I've ever implied that exposing more to the right will not provide more levels. I use the evaluative metering mode quite often. At ISO 100 I notice that quite often with low DR subjects I could reshoot the scene with an additional stop of exposure. Doing so gives an over all appearance of an overexposed shot and introduces a risk of inadvertantly blowing wanted detail in highlights. I tend to not want to take that risk for the sake of perhaps smoother tonality of pixel-peeping proportions.
But you are of course quite right that more levels are better than fewer levels. I would not argue against that. At higher ISOs, I get more anxious about ETTR because I know from experience that a difference of just 1/2 a stop of exposure can make a worthwhile difference to shadow and mid tone noise. At lower ISOs I feel I can be more relaxed about the need for ETTR. At ISO 100 with a low DR subject, I can get so relaxed I might not even bother to reshoot the scene if there's not much showing on the left side of the histogram.
-
Again, I must point out the importance of determining your sensor's exact ISO and dynamic range and to expose to just retain textural detail without clipping. But without certain knowledge of your sensor's ISO and dynamic range, most people fail to actually use that densely packed area of near clipped data.
[a href=\"index.php?act=findpost&pid=62676\"][{POST_SNAPBACK}][/a]
A very interesting and needed thread. I don't own a digital camera, and have always wondered how its sensor differ from film, but have never found a book or article that clearly explains it. If such literature exist, please point me to them.
At the risk of adding more confusion, here's one book author's attempt (he may be too simplistic, but at least he tried):
"The digital equivalent of the exposure-latitude variations between slide and negative is the selection of file format. Shooting RAW files is comparable to exposing negative film. The unprocessed file allows images that were under or overexposed by up to two, or even three, stops to be successfully converted and printed. TIFF files also offer a fair bit of latitude, with over- and underexposure by up to 1-1/2 stops not causing too many problems. Shooting JPEGs is more like exposing slide film. The processed file demands greater accuracy at the time of capture, but excellent results can still be achieved with files over and underexposed by one stop."
BTW, how can one determine a "sensor's exact ISO and dynamic range"?
-
If you compare a 'correctly' exposed shot at ISO 1600 (exposed fully to the right) with the same shot at ISO 100 (same shutter speed and f stop), the ISO 100 shot will appear to be underexposed. However, the sensor has received the same amount of light, yet shadow noise will be very much greater. This implies that noise reduction cna be more successfully implemented when the signal is boosted.
[a href=\"index.php?act=findpost&pid=62906\"][{POST_SNAPBACK}][/a]
Indeed, and it is not just a matter of noise reduction: underexposing by four stops at ISO 100 means that the signal passes through the last pre-amplifier stages 1/16th as strong, and then has to be amplified 16 times as much in the digital domain to get correct levels. This means that an noise introduced late on the pre-amplifier stage and "quantization noise" from A/D conversion gets amplified by an extra factor of 16.
Part of optimal choice of the three exposure parameters (aperture, shutter speed and ISO speed) to minimize shadow noise is amplifying the signal sent into the A/D converter as much as possible without having the highlight signal stronger than the A/D convertor can handle. So even at high ISO speeds, where the sensor is getting well below maximum exposure, you might want to push the "A/D convertor input histogram" to the right.
For example, for a scene with details of interest in rather deep shadows and highlights not going far above the mid-tones, it might be better to "overexpose" by one stop at ISO 800 vs "on meter" exposure at ISO 400 (so same shutter speed and aperture) if the system has enough highlight headroom to handle this overexposure. But you do then have to fiddle with reducing brightness in PP though, pushing the tone curve down in the mid-tones or something like that.
This is a bit like "exposing for the shadows" when the shadows are of interest, while keeping the highlights just within gamut.
-
A very interesting and needed thread. I don't own a digital camera, and have always wondered how its sensor differ from film, but have never found a book or article that clearly explains it. If such literature exist, please point me to them.
At the risk of adding more confusion, here's one book author's attempt (he may be too simplistic, but at least he tried):
"The digital equivalent of the exposure-latitude variations between slide and negative is the selection of file format. Shooting RAW files is comparable to exposing negative film. The unprocessed file allows images that were under or overexposed by up to two, or even three, stops to be successfully converted and printed. TIFF files also offer a fair bit of latitude, with over- and underexposure by up to 1-1/2 stops not causing too many problems. Shooting JPEGs is more like exposing slide film. The processed file demands greater accuracy at the time of capture, but excellent results can still be achieved with files over and underexposed by one stop."
BTW, how can one determine a "sensor's exact ISO and dynamic range"?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=62911\")
I don't think many of the forum experts would agree with your reference on exposure. With raw files, one can recover one half to one stop of overexposure, but with more overexposure the highlights will be clipped and data irretrievably lost. With significant underexposure, dynamic range and noise will suffer, expecially when shooting at higher ISOs.
Here is an explanation of sensor ISO as it applies to digital sensors:
[a href=\"http://www.normankoren.com/digital_cameras.html#ISOspeed]http://www.normankoren.com/digital_cameras.html#ISOspeed[/url]
If you take a picture of a uniformly lit gray or white card (it doesn't make any difference), and render the image in sRGB or aRGB, you should get a pixel value of 114 if your system is properly calibrated.
Dynamic range is defined by electronic engineers as linear full well (electrons)/read noise (electrons). This is discussed by Roger Clark:
http://www.clarkvision.com/imagedetail/eva...-1d2/index.html (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html)
Since the floor of dynamic range is determined by noise, the DR is much lower at high ISOs than lower ISOs as shown in Roger's tables. The presentation is rather technical one can get most of the important information merely by inspecting the tables.
-
I use the evaluative metering mode quite often. At ISO 100 I notice that quite often with low DR subjects I could reshoot the scene with an additional stop of exposure. Doing so gives an over all appearance of an overexposed shot and introduces a risk of inadvertantly blowing wanted detail in highlights. I tend to not want to take that risk for the sake of perhaps smoother tonality of pixel-peeping proportions.
[a href=\"index.php?act=findpost&pid=62908\"][{POST_SNAPBACK}][/a]
Ray,
Your approach makes sense. In a previous thread on the Adobe forum I engaged in a discussion with Jeff and Bruce Fraser about the dangers of highlight loss, but with more experience, study, and testing I have come closer to embracing their views. With reasonably careful exposure and with highlight recovery in ACR the dangers of high light loss are minimal.
That said, shooting at ISO 100 rather than 400 can improve DR and shadow noise more than ETTR unless you are really under exposing.
-
I'm still travelling so don't often get on the net. I see you've misunderstood what I mean here. Exposure is determined by shutter speed and f stop. No matter what the ISO, if the exposure is the same, then the same amount of light falls upon the sensor. An ISO setting is merely an instruction to the camera to amplify the signal and apply certain noise reduction processes. If you compare a 'correctly' exposed shot at ISO 1600 (exposed fully to the right) with the same shot at ISO 100 (same shutter speed and f stop), the ISO 100 shot will appear to be underexposed. However, the sensor has received the same amount of light, yet shadow noise will be very much greater. This implies that noise reduction cna be more successfully implemented when the signal is boosted.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=62906\")
Yes, I did mis-read your original post. I hardly expected that anyone would give an ISO 1600 exposure with the camera set at ISO 100. Why would you do that? However, if you are shooting in RAW I'm not sure that the noise would be that much different. Have you done the experiment? The main determinant of noise is photon counting statistics and the same number of photons would be captured in both cases. However, without the additional signal amplification, the analog to digital conveter would produce many fewer levels with the ISO 100 setting and there would be poor quantification.
When recording very low light levels, such in astrophotography, Roger Clark reports that with the Canon EOS 1D Mark II, it is better to use ISO 1600 than 3200, presumably using exposure adjustment in rendering the raw image so as to make better advantage of the read noise at ISO 1600.
[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]
-
I hardly expected that anyone would give an ISO 1600 exposure with the camera set at ISO 100. Why would you do that?
Two reasons. First, just plain curiosity, and second a real practical reason. To change ISO on the D60, my first DSLR, I had to scroll through a menu selection. There were times when I needed to take a shot in a hurry to capture the moment. Stuffing around with ISO selection was not on. If my camera was set on ISO 100, which it was most of the time, I needed to know if an underexposed ISO 100 shot was significantly more noisy than a correctly exposed ISO 400 shot. With the D60 it wasn't. However, the D60 didn't extend to ISO 1600.
Have you done the experiment? The main determinant of noise is photon counting statistics and the same number of photons would be captured in both cases.
I've just repeated the experiment with my 5D. I arrived back in Brisbane this morning. I haven't had the opportunity yet to process the thousands of shots of temples in the jungles of Cambodia, Hill tribes in the north of Vietnam and elephants squirting water over people during the Songkran festival in Thailand, etc etc., but shooting a few test shots from my back yard to demonstrate a point is easy, thanks to the digital revolution.
The first is the full scene taken at dusk; ISO 1600, 100th sec, f8, exposed fully to the right.
The next two are 100% crops of the lower left corner of that scene at ISO 100 and ISO 1600. I think there's no need to label them. The ISO 100 shot is very much noisier.
The settings for conversion in ACR were; shadows zero; contrast zero; brightness default 50. For the ISO 1600 shot I used minus 1 EC and the ISO 100 shot plus 3 EC, the difference being 4 stops of EC between the 2 exposures.
[attachment=486:attachment]
[attachment=487:attachment]
[attachment=488:attachment]
ps. Forgot to mention - Canon 24-105 IS at 24mm.
-
The first is the full scene taken at dusk; ISO 1600, 100th sec, f8, exposed fully to the right.
The next two are 100% crops of the lower left corner of that scene at ISO 100 and ISO 1600. I think there's no need to label them. The ISO 100 shot is very much noisier.
The settings for conversion in ACR were; shadows zero; contrast zero; brightness default 50. For the ISO 1600 shot I used minus 1 EC and the ISO 100 shot plus 3 EC, the difference being 4 stops of EC between the 2 exposures.
ps. Forgot to mention - Canon 24-105 IS at 24mm.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63062\")
Ron,
Very interesting experiment! The noise at the ISO 1600 does have markedly less noise. Since the exposure is the same, the amount of light hitting the sensor and the photon noise should be the same in both cases.
According to Roger Clark's analysis of noise on the EOS 1D Mark II, the main sources of noise are photon noise and read noise. Dark current enters the equation at very low illuminance levels. According to Roger's tests, the read noise of the above camera is 3.90 electrons at ISO 1600 and 16.61 at ISO 100. Perhaps the lower read noise at ISO 1600 is responsible for the lower noise without invoking increased NR at 1600.
Ron Parr, who is a computer science professor at Duke and an avid photographer, states that he sees no evidence that Canon applies additional NR at ISO 1600:
[a href=\"http://forums.dpreview.com/forums/read.asp?forum=1021&message=18059253]http://forums.dpreview.com/forums/read.asp...essage=18059253[/url]
-
According to Roger's tests, the read noise of the above camera is 3.90 electrons at ISO 1600 and 16.61 at ISO 100. Perhaps the lower read noise at ISO 1600 is responsible for the lower noise without invoking increased NR at 1600.
Ron Parr, who is a computer science professor at Duke and an avid photographer, states that he sees no evidence that Canon applies additional NR at ISO 1600:
The above 2 statements are contradictory. Since ISO 1600 shots are clearly less noisy than ISO 100 shots when equal amounts of light fall on the sensor, then additional noise reduction must have been applied, either actively or by default.
As I understand, the major component of readout noise arises from the on-chip pre-amplifier. The greater the amplification (prior to A/D conversion) the greater the read noise. If Roger Clark has tested a lower read noise at ISO 1600 than at ISO 100, then that would be proof that additional noise reduction has been applied at ISO 1600. (Assuming Roger's methodology is sound).
I repeated the above experiment at mid-day to check if the noise differences are as great in better lighting conditions. They are. Exposure was 1/200th at f22 for both shots. During RAW conversion I applied -0.8 EC to the 1600 image and +3.2 EC to the ISO 100 image. The ISO 1600 image has better (smoother) tonality in all parts of the image, including the water and the sky.
[attachment=496:attachment] [attachment=497:attachment]
-
it might be better to "overexpose" by one stop at ISO 800 vs "on meter" exposure at ISO 400 (so same shutter speed and aperture) if the system has enough highlight headroom to handle this overexposure.
BJL,
This appears to be the case. However, if the system has enough highlight headroom to fully expose to the right at ISO 800, then there's enough headroom to give double the exposure at ISO 400, which of course will produce even better tonality in the shadows.
-
The above 2 statements are contradictory. Since ISO 1600 shots are clearly less noisy than ISO 100 shots when equal amounts of light fall on the sensor, then additional noise reduction must have been applied, either actively or by default.
As I understand, the major component of readout noise arises from the on-chip pre-amplifier. The greater the amplification (prior to A/D conversion) the greater the read noise. If Roger Clark has tested a lower read noise at ISO 1600 than at ISO 100, then that would be proof that additional noise reduction has been applied at ISO 1600. (Assuming Roger's methodology is sound).
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63145\")
These statements are hardly contradictory at all. You have it backwards: read noise is lower at high ISO than low ISO. As discussed in the Kodak white paper below, read noise has two main components: white noise and flicker noise. In the equation for white noise, the amplifier gain is in the denominator and hence greater gain (used at higher ISO) results in less white noise. The flicker noise has to do with with speed of the readout--it is greater when the pixels are read out faster. The discussion is for CCD, but the same principles apply to CMOS (according to Ron Parr).
[a href=\"http://www.kodak.com/global/plugins/acrobat/en/digital/ccd/applicationNotes/noiseSources.pdf]http://www.kodak.com/global/plugins/acroba...oiseSources.pdf[/url]
The read noise can be determined by simple user tests as Roger describes. I have determined the read noise for the Nikon D200. It is about 14.3 electrons at ISO 100 and 9.9 electrons at ISO 1600. Roger's values for the Canon are 16.61 and 3.93 electrons respectively. The Nikon does no on chip NR, so your theory does not seem to be correct. Your observations are correct, but I suspect that your interpretation is not.
The higher read noise of the Nikon D200 at ISO 1600 as compared to the Canon EOS 1D Mark II (and most likely other Canon cameras) may explain at least in part why the Canons have better high ISO performance. Read noise has a profound effect with low signal levels, whereas photon sampling noise predominates at higher signal levels.
-
BJL,
This appears to be the case. However, if the system has enough highlight headroom to fully expose to the right at ISO 800, then there's enough headroom to give double the exposure at ISO 400, which of course will produce even better tonality in the shadows.
[a href=\"index.php?act=findpost&pid=63146\"][{POST_SNAPBACK}][/a]
I am talking about giving the sensor maximum possible exposure before highlights bloom. If a certain exposure level (in the sense of shutter speed and aperture choice) just avoids overflow of highlight pixels, it does that regardless of ISO setting: for example if ISO 800 "on meter" fills some pixels to just below maximum, then so will ISO 400 one stop under, while ISO 400 "on meter" will involve double the exposure (by halving shutter speed or opening up one stop larger aperture or whatever) which will blow out some highlights. So there will then be a cost to the improved shadow handling.
My proposed guideline:
1. choose exposure level (shutter speed and f-stop) to give maximum exposure of the sensor, and then
2. choose the highest ISO setting that does not pre-amplify the sensor's signal so much that highlights are clipped in A/D conversion.
Typically step 2. will mean using the ISO speed that goes with "on meter" exposure, but a bit under for subjects with a wide range from mid-tones to highlights, and maybe a bit over for subjects with a narrow range there.
-
I am talking about giving the sensor maximum possible exposure before highlights bloom. If a certain exposure level (in the sense of shutter speed and aperture choice) just avoids overflow of highlight pixels, it does that regardless of ISO setting: for example if ISO 800 "on meter" fills some pixels to just below maximum, then so will ISO 400 one stop under, while ISO 400 "on meter" will involve double the exposure (by halving shutter speed or opening up one stop larger aperture or whatever) which will blow out some highlights. So there will then be a cost to the improved shadow handling.
My proposed guideline:
1. choose exposure level (shutter speed and f-stop) to give maximum exposure of the sensor, and then
2. choose the highest ISO setting that does not pre-amplify the sensor's signal so much that highlights are clipped in A/D conversion.
Typically step 2. will mean using the ISO speed that goes with "on meter" exposure, but a bit under for subjects with a wide range from mid-tones to highlights, and maybe a bit over for subjects with a narrow range there.
[a href=\"index.php?act=findpost&pid=63183\"][{POST_SNAPBACK}][/a]
Still can't follow your reasoning here, BJL. It's possible to give the sensor maximum possible exposure, in terms of real photons, only at base ISO. At higher than base ISO, the well will never be full using a 'correct' or metered exposure for that higher ISO. If you are going to follow the procedure of exposing fully to the right, short of clipping highlights and consistent with an appropriate exposure/f stop combination for your intentions, then doing so at the lowest ISO possible will always produce better tonality.
If a metered exposure at, say, ISO 400 looks as though it could be a stop greater, then sticking with ISO 400 and doubling the exposure will produce better results than moving up to ISO 800 using the metered exposure for ISO 400, provided the slower shutter speed is still fast enough for the conditions.
-
It's possible to give the sensor maximum possible exposure, in terms of real photons, only at base ISO.
[a href=\"index.php?act=findpost&pid=63203\"][{POST_SNAPBACK}][/a]
No: the amount of light delivered to each photosite is determined solely by aperture an shutter speed: ISO speed setting only effects the subsequent processing during read-out of the electrons gathered in the photosites. If f/16 1/100s ISO 100 in bright sun fills the highlight photosites, then so does f/16 1/100s, ISO 200, but in the latter case, that signal is amplified twice as much before A/D conversion, which might cause clipping in the amplifier or A/D convertor.
At higher than base ISO, the well will never be full using a 'correct' or metered exposure ...
[a href=\"index.php?act=findpost&pid=63203\"][{POST_SNAPBACK}][/a]
Aha, we are talking about different cases. I was explicitly _not_ talking about "correct metered exposure", but equal exposure levels, meaning equal shutter speed and f-stop. My scenario is a kind of full manual mode, selecting shutter speed and aperture (to fill highlight photosites) and then also choosing ISO speed. Normally, my procedure would end up choosing the ISO that gives "correct metered exposure", but sometimes, a bit of overexposure (when highlights can stand it) could help with shadow noise.
-
These statements are hardly contradictory at all. You have it backwards: read noise is lower at high ISO than low ISO. As discussed in the Kodak white paper below, read noise has two main components: white noise and flicker noise. In the equation for white noise, the amplifier gain is in the denominator and hence greater gain (used at higher ISO) results in less white noise. The flicker noise has to do with with speed of the readout--it is greater when the pixels are read out faster. The discussion is for CCD, but the same principles apply to CMOS (according to Ron Parr).
Maybe I have got it backwards. There's a lot of misinformation on the net. Ultimately, one can only go along with what makes sense and what gels with other concepts one thinks one has understood. The idea that a signal can be amplified with a result that noise is less in absolute terms rather than relative terms does not make sense to me, in the absensce of a noise reduction technique. Yet the formula you refer to in the Kodak paper certainly suggest that this is the case. I'm not a mathematician but I cans see that a numerator enclosed in a square root is not going to escalate in value greatly and a denominator consisting of what is essentially a constant (efficiency) multiplied by a gain figure, can only have an effect of reducing the over all value as gain increases.
Having just looked at Roger's table of measurements you've provided a link to, it's clear he is not measuring the readout noise of equal photon counts at different ISOs. As ISO increases, the absolute value of readout noise is shown as falling, but dynamic range and signal-to-noise falls much more dramatically. As you can see from my example images, in situations where the photon count is the same, the higher ISO image has better S/N and DR.
What appears to be happening here (it's the only interpretation I can think of) is the signal being read is approximately the same level at all ISOs. The difference is, the signal at lower ISOs has resulted from a higher photon count combined with lower gain. It sort of makes sense that a signal that has been amplified in a controlled fashion might be easier to read than a signal that is just stronger initially.
Nevertheless, I still find it difficult to accept that there is no additional noise reduction going on at high ISOs. If one looks at the history of Canon DSLR development, there appears to be no dramatic improvement in S/N and DR at base ISO. The 1Ds actually had slightly worse noise than the D60 at ISO 100, on a pixel for pixel basis. Nor did the later 10D have better noise characteristics than the D60 at ISO 100, but it certainly did at ISO 400 and above, and this improvement at high ISOs has continued with the 20D and 5D, so it seems clear to me that that simple formula you refer to in the Kodak paper is not telling the whole story. I would also find it difficult to believe that that equation represents any recent development in preamplifier and/or CMOS chip design.
Whilst doing a bit of research on Google, I came across an interesting thread on Photonet, which addresses some of these issues. The article here (http://www.photo.net/learn/dark_noise/) is mainly about dark frame subtraction during post processing, which doesn't appear to produce consistent results because this sort of thing is best done in-camera prior to demosaicing. However, the author, Jeff Medkeff, seems convinced that modern CMOS imagers have on-chip noise reduction devices for readout noise which he refers to as 'bias noise'.
Following are a couple of relevant quotes from his article.
Bias noise is also highly repeatable - but since it is a result of reading out the sensor, it does not even depend on shooting conditions being the same. Practically the only variable affecting readout noise in a digital camera exposure is the amount of amplifier gain. As long as the amplifier gain remains the same, readout noise will be nearly identical from shot to shot. In general, doubling amplifier gain can be expected to approximately double the amount of readout noise.
Is Jeff dead wrong here?
CMOS sensors allow the placement of both photosites and transistors on the sensor itself. (CCDs cannot have any processing circuitry built into the sensor - just transfer gates and the like, which are controlled by off-sensor control circuitry.) Because of this, CMOS sensors generally have at least the readout amplifier built in to the photosite. There may be other transistors as well, which perform other processing steps. It is now very common for a CMOS sensor to include noise-reduction circuitry directly on the sensor alongside the readout amplifier. In some designs, a sort of small dummy photosite, shaded from light, is used to quantify the likely dark noise level in the actual photosite, and this quantity is subtracted during readout. In other designs, a constant - corresponding to the tested dark current of the sensor - is subtracted from the photosite value during readout. If anything like this is happening, expectations such as "dark noise will double with twice the exposure duration" may turn out to be false.
This makes sense to me but what about the next statement?
In addition, this on-sensor circuitry can be designed to subtract the amount of bias noise that the sensor designer expects will be contributed to that particular pixel. This is a design-time decision, so bias noise may still be introduced due to manufacturing variations, erroneous expectations on the part of the designer, changes in other circuitry at a later point in development that the designer decided not to compensate for, and so forth. In any case, if bias noise is being addressed in a CMOS sensor camera - and it is being aggressively dealt with in all known current DSLRs - the relationship between ISO and readout noise in a particular camera's images might not be as simple or as repeatable as expected.
Finally, there's one factor which might have a much greater effect (than lower read noise) on noise reduction of equal initial signals at high ISOs, and that is the number of bits available to describe the signal at the A/D conversion stage. Whether the signal is pushed to the right as a result of greater photon count or greater in-camera amplification, it's pushed to the right nevertheless and more levels are available during digitisation.
-
Aha, we are talking about different cases. I was explicitly _not_ talking about "correct metered exposure", but equal exposure levels, meaning equal shutter speed and f-stop.
BJL,
My last few posts and example images are about this exact scenario. Higher ISOs result in cleaner images using the same exposure. But the point which I think you are obscuring is that lower ISOs with appropriately greater exposure result in even cleaner images.
As bjanes asked in a previous post, why would anyone want to deliberately underexpose at a lower ISO setting? The only practical reason for doing so would be the possibility of losing a shot because of the time taken to move up to a higher ISO. In order to determine whether or not additional exposure (or additional preamplifier gain) will blow highlights, it's necessary to have a metered reading so I don't understand your comment that you are not referring to a 'correct metered exposure' scenario, unless you are quibbling about the use of the term 'correct' which I deliberately place in quotes to indicate it's a loose term.
Let's put it this way. Whichever way you look at it, an accurate exposure reading is necessary whether it's a reading from the camera's meter, an external meter, or a histogram, or an educated guess. Having determined how much highlight headroom is available from an accurate metered reading of one sort or another, it then makes more sense to increase exposure accordingly at the lowest ISO possible, consistent with an adequately fast shutter speed for the conditions. If use of a low ISO setting results in an inadequate shutter speed for an ETTR situation, or a DoF which is too shallow, then it's advantageous to move up to a higher ISO setting.
-
Maybe I have got it backwards. There's a lot of misinformation on the net. Ultimately, one can only go along with what makes sense and what gels with other concepts one thinks one has understood.
Having just looked at Roger's table of measurements you've provided a link to, it's clear he is not measuring the readout noise of equal photon counts at different ISOs. As ISO increases, the absolute value of readout noise is shown as falling, but dynamic range and signal-to-noise falls much more dramatically. As you can see from my example images, in situations where the photon count is the same, the higher ISO image has better S/N and DR.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63216\")
Yes, there is a lot of misinformation on the web and you must carefully evaluate what you read there. If you look at Roger's bio
[a href=\"http://www.clarkvision.com/rnc/index.html]http://www.clarkvision.com/rnc/index.html[/url]
you will see he is highly qualified in this area, having received a PhD in astrophysics from MIT, published more than 160 sceintific papers, and participated in many NASA imaging projects. He does not measure readout noise at equal photon counts because readout noise is not correlated with the photon count, but remains constant for a given ISO. If you read his methodolgy, you will see that he determines read noise with the lens cap on, resulting in a very low photon count.
Following are a couple of relevant quotes from his article.
In general, doubling amplifier gain can be expected to approximately double the amount of readout noise.
Is Jeff dead wrong here?
[a href=\"index.php?act=findpost&pid=63216\"][{POST_SNAPBACK}][/a]
Most likely he is, since his assertions do not agree with Roger's or the Kodak white paper. I note that he is a photographer and amateur astronomer and not an imaging scientist. If you look at the Kodak white paper on white noise, you will see that neither the numerator nor the denominator contains any reference to the photon count. Since amplifier gain is in the denominator, increased gain should decrease white noise, consistent with Roger's article.
Furthermore, if you read Roger's article, you will see that he measures noise observationally, and then calculates photon noise, read noise, and dark noise and verifies that the calculations match the observed noise, thereby double checking the results.
In practice, one does not hold the photon count constant, but rather as photon count decreases, the amplifier gain is increased so that the resultant voltage matches the scale of the AD converter. I do not understand your obsession with holding the photon count constant--that is not what is done in practice.
I must admit that I am not am imaging scientist either, and only learned about most of these topics recently after having read Roger's posts. Hopefully a real scientist will jump into the thread and clarify things.
-
you will see he is highly qualified in this area, having received a PhD in astrophysics from MIT, published more than 160 sceintific papers, and participated in many NASA imaging projects.
That may be so, but I've long since stopped accepting the word of people just because they appear to be exceptionally qualified, because I'm aware there is a long list of exceptionally learned and intelligent people from the time of Aristotle and beyond who have been dead wrong on a number of issues, in retrospect of course. But I will read more of Roger's articles when I have time.
He does not measure readout noise at equal photon counts because readout noise is not correlated with the photon count, but remains constant for a given ISO. If you read his methodolgy, you will see that he determines read noise with the lens cap on, resulting in a very low photon count.
You're contradicting yourself here, bjanes. Whether it's a low photon count or even a zero photon count, if the lens cap is on it's an equal photon count. If this is the case, then the fewer electrons of noise at ISO 1600 could not explain the dramatic reduction of noise I see at ISO 1600.
I do not understand your obsession with holding the photon count constant--that is not what is done in practice.
No obsession. Just trying to get a handle on possible future developments. The phenomenon of being able to clean up noise so dramatically simply by amplifying the signal causes me to wonder if the reverse could be done simultaneously on the right side of the histogram, ie. compressing the higher tones to achieve greater dynamic range. I've spent a lot on DSLRs (a lot by my standards). I have a D60, a 20D and a 5D and each of these camera's I've bought as result of either significant resolution improvement and/or dramatic noise reduction at high ISOs. I hope this trend continues and since it's going to affect my wallet if it does continue, I have a fairly strong interest in possible trends and improvements.
Cheers!
-
That may be so, but I've long since stopped accepting the word of people just because they appear to be exceptionally qualified, because I'm aware there is a long list of exceptionally learned and intelligent people from the time of Aristotle and beyond who have been dead wrong on a number of issues, in retrospect of course. But I will read more of Roger's articles when I have time.
[a href=\"index.php?act=findpost&pid=63264\"][{POST_SNAPBACK}][/a]
Ray, you appear to accept whatever agrees with your preconceived notion, and ignore what does not agree with your preconception. It is impossible to reason with you and I will end my responses to you at this time.
You're contradicting yourself here, bjanes. Whether it's a low photon count or even a zero photon count, if the lens cap is on it's an equal photon count. If this is the case, then the fewer electrons of noise at ISO 1600 could not explain the dramatic reduction of noise I see at ISO 1600.
[a href=\"index.php?act=findpost&pid=63264\"][{POST_SNAPBACK}][/a]
This is the second time I have contradicted myself (according to you), without actually having done so. The lens cap is in place so that the observed noise is due to read noise (and a very small amount of dark current) and not to Poisson photon noise. The purpose of the lens cap is not to hold the photon count constant.
Goodbye
-
BJL,
My last few posts and example images are about this exact scenario. Higher ISOs result in cleaner images using the same exposure.
[a href=\"index.php?act=findpost&pid=63225\"][{POST_SNAPBACK}][/a]
We agree!
But the point which I think you are obscuring is that lower ISOs with appropriately greater exposure result in even cleaner images.
[a href=\"index.php?act=findpost&pid=63225\"][{POST_SNAPBACK}][/a]
I am not obscuring that at all; I repeat, we are simply considering different situations.
You are now considering the case where one has the option of increasing sensor exposure, in the sense of a longer exposure time or a larger aperture. Then it is not surprising that working at low "base ISO" speed is best.
I was considering the case you previously raised with your ISO 1600 vs "ISO 100 four stops underexposed": when light is scarce so that even with maximum possible sensor exposure there is no chance of any photosites getting "full" and using base ISO would involve underexposure. Then the best procedure seems to be to raise the ISO (pre-amplification before A/D conversion) so that the A/D convertor sees a "full strength" signal, with the brightest highlights producing near maximum levels in the RAW digital output. Probably often "on meter", but maybe an even higher ISO and so a bit "over meter" when the highlights do no go far above the metered mid-tones and so the mid-tones (and shadows) can be pushed further right on the histogram.
Note on another discussion in this thread: be careful to distinguished absolute levels of noise from signal-to-noise ratios. Increased pre-amplification will likely increase absolute noise levels, but sometimes by less than the increase in signal level, resulting in increased signal-to-noise ratio, often described as reduced noise.
My rough explanation is that some noise is introduced after some or all pre-amplification has happened, and so if the signal is stronger when that noise enters (due to greater pre-amplification), the ratio of that signal to the new noise is higher (better).
To put it another way: you need the same amount of amplification in then end, including adjusting levels up in an underexposed image, so it is always better to have part of the noise enter after part of the amplification has been done rather than before, to minimize further amplification of that part of the noise.
-
I don't understand your comment that you are not referring to a 'correct metered exposure' scenario
[a href=\"index.php?act=findpost&pid=63225\"][{POST_SNAPBACK}][/a]
Of course we want "correct" exposure, but my original phrase was "on meter", not "correct".
Perhaps all I am saying is that in some high ISO speed situations where shutter speed and aperture are constrained so that only ISO speed can be varied, the "correct" choice of ISO is not always the one that the meter confirms. It can instead be a higher one which the meter will call overexposed (when on meter exposure wastes highlight headroom), or a lower one which the meter will call underexposed (when "on meter" exposure causes highlight clipping in pre-amplification or A/D conversion.)
Which is similar to advice about choosing a suitably high speed film, with the digital version in some cases effectively using ISO 800 film but rated at EI 400 (to hold shadows better), or ISO 400 film rated at EI 800 (to hold a strong highlights).
-
I was considering the case you previously raised with your ISO 1600 vs "ISO 100 four stops underexposed": when light is scarce so that even with maximum possible sensor exposure there is no chance of any photosites getting "full" and using base ISO would involve underexposure. Then the best procedure seems to be to raise the ISO (pre-amplification before A/D conversion) so that the A/D convertor sees a "full strength" signal, with the brightest highlights producing near maximum levels in the RAW digital output. Probably often "on meter", but maybe an even higher ISO and so a bit "over meter" when the highlights do no go far above the metered mid-tones and so the mid-tones (and shadows) can be pushed further right on the histogram.
BJL,
I still can't follow the point you are making, unless it's the obvious one that a higher ISO is useful when the shutter speed at a lower ISO is too long. The first 2 shots I took at dusk were at 100th and f8 using an IS lens. The 4 stops of underexposure at ISO 100 could have been corrected by using 1/6th sec at the same aperture. This would have resulted in some of the photosites (relating to the sky) getting full. Surely the situations, where light is so scarce that maximum possible sensor exposure will not fill any of the photosites, are very rare. Some possible examples might be night photography with no artificial lighting or moon, or taking a shot with the lens cap on.
However, 1/6th of a sec is too long for a hand-held shot, even with the benefit of IS and a short focal length. F8 was also giving me too shallow a DoF. If those shots were intended to be more than experiments, I would have used a tripod, MLU, 1/3rd sec exposure at f11 and ISO 100. If I didn't have a tripod handy, I would probably have taken a shot at 1/25th, f11 and ISO 800 and another at 1/50th, f11 and ISO 1600, just in case the first was not acceptably sharp.
All these exposures would result in a small amount of highlight-warning flashing in the sky (observed in the shot at 100th, f11 and ISO 1600) indicative of a full exposure to the right. It was necessary to use -1 EC during conversion to recover the small degree of blown highlight. If the sky had been overcast and grey, I could probably have given an extra 1/3 to 1/2 stop of 'overexposure'.
I still can't see any advantage to using a higher ISO other than the opportunity to use a faster shutter speed and/or smaller aperture, but perhaps a little more experimentation is required .
-
BJL,
I still can't follow the point you are making, unless it's the obvious one that a higher ISO is useful when the shutter speed at a lower ISO is too long. .
[a href=\"index.php?act=findpost&pid=63324\"][{POST_SNAPBACK}][/a]
A possible key: "pre-amplifier" or "read" noise and "quantization" noise in A/D convertors are factors too, not only photon ("shot") noise and sensor dark current noise.
It should be clear that I am not talking about using higher ISO to get a higher shutter speed: I am talking about the very case that you first raised, of using a higher ISO with the same shutter speed and same f-stop, perhaps to the point that the light meter says that you are overexposing, and the straight output looks too bright, but highlights are not blown-out and you can restore desired luminosity with PP like minus compensation.
The point of this is that it takes the same signal (and same photon and dark current noise) from each photosite and pre-amplifies it more before A/D conversion, which reduces the final effect of noise that enters "midstream": quantization noise in the A/D converter and read noise in the analogue steps just before A/D. This benefit is because that equal amount of midstream noise is added to a signal that is already stronger, due to having already been amplified more. Since this stronger signal produces higher levels in A/D output, it needs lower subsequent amplification in the digital domain to get a given luminosity level in the final displayed image. And thus the mid-stream noise gets amplified less in the digital domain, and comes out weaker in the final output and the final displayed image.
On the other hand, noise already present on the sensor comes out at the same level either way. So if you think only in terms of photon and dark current noise, there would seem to be no difference.
Aside: Dalsa spec sheets show that increasing read speed (on their Full Frame CCD's at least) can about double the total dark noise, = sensor dark current noise+read noise. So read noise can be quite significant, at least if processing has to be done fast. Also, excessively slow read-out can increase total dark noise: apparently letting the signal to sit around too long in the electron wells allows additional dark current noise to accumulate.
BJL,
I still can't see any advantage to using a higher ISO other than the opportunity to use a faster shutter speed and/or smaller aperture, but perhaps a little more experimentation is required[a href=\"index.php?act=findpost&pid=63324\"][{POST_SNAPBACK}][/a]
You already did an extreme version of this experiment and confirmed what I am saying: ISO 100 underexposed by four stops versus ISO 1600 "on meter"!
-
You already did an extreme version of this experiment and confirmed what I am saying: ISO 100 underexposed by four stops versus ISO 1600 "on meter"!
[a href=\"index.php?act=findpost&pid=63400\"][{POST_SNAPBACK}][/a]
BJL,
That's true. I've already confirmed that adjustable ISO serves a real and worthwhile purpose (apart from producing a brighter review on the camera's LCD screen). But the experiment has not confirmed the point you are making, which I now understand to be as follows.
In the case where, say, one stop of highlight headroom is available at a certain exposure at, say, ISO 100, there would be essentially 2 ways of exposing fully to the right to use that headroom. One is to halve the shutter speed using the same ISO setting of 100. The other is to keep the same exposure but jump up to ISO 200.
I have no doubt that the latter option is preferrable to using the same exposure at ISO 100 (which would essentially be 1 stop underexposure for ETTR purposes). But I do have doubts that the latter option would produce better tonality than the first option as described (ie ISO 100 and half the shutter speed).
In other words, if there's a contest between double the photon count and double the preamplification gain, then I'd expect that double the photon count will produce better tonality however marginal it may be. It's this factor I haven't yet tested, but you've put me in a position where I now feel that I need to test it.
If I'm wrong, I'll eat my hat .
-
One is to halve the shutter speed using the same ISO setting of 100. The other is to keep the same exposure but jump up to ISO 200.
... if there's a contest between double the photon count ...
Ray, you still do not get what I am saying, even after I used italics in my previous post! There is no contest involving double the photon count. I am comparing different ISO speed setting options that involve exactly the same photon count at each photosite due to using exactly the same shutter speed and aperture, but different amounts of pre-amplification so that those same photon counts are converted to different voltages being input to the A/D converter, giving different digital levels going into the RAW output. My goal is "exposing the A/D convertor to the right", giving it the highest input voltage that it can handle so that the brightest parts of the image are represented in the RAW file by near-maximum digital levels.
And this is specifically about situations calling for higher than base ISO speed in order to get "on meter" exposure, so it is best to stay away from examples involving ISO 100!
-
Of course we want "correct" exposure, but my original phrase was "on meter", not "correct".
Which is similar to advice about choosing a suitably high speed film, with the digital version in some cases effectively using ISO 800 film but rated at EI 400 (to hold shadows better), or ISO 400 film rated at EI 800 (to hold a strong highlights).
[a href=\"index.php?act=findpost&pid=63282\"][{POST_SNAPBACK}][/a]
The above works well for film with exposure determined from middle gray. However, with ETTR in digital, one should place the highlight just short of clipping (data value 4095 in the raw file with a 12 bit AD converter). The easiest way to do this is probably to take a spot reading from the highlight and expose according to a previously determined offset from the nominal meter reading that places the highlights properly. This is often about 2 1/3 stops over the indicated exposure.
For the best signal to noise, one should use the lowest ISO that is feasable for the shooting conditions. It is hardly surprising that in dim light a high ISO may be needed. The amplifier gain is then increased to compensate for the reduced exposure, so that the highlights will be at the desired location in the histogram. If one uses an inappropriately low ISO for the exposure, underflow of the AD converter may occur, shadow values will be lost and there will be a reduced number of tones at higer signal levels. In such cases, one has not really exposed to the right.
With JPEG, some cameras may use different noise reduction for high ISO, but this is not well documented with raw. Ron Parr (who publishes a digital FAQ at Duke Univesity), has stated that he thinks that, at high ISO, Canon merely reduces the sharpening and he has seen no evidence that additional NR is given. In the case of Nikon, the Nikon guru Thom Hogan has stated that Nikon applies no NR to the raw image.
-
Ray, you still do not get what I am saying, even after I used italics in my previous post! There is no contest involving double the photon count. [a href=\"index.php?act=findpost&pid=63486\"][{POST_SNAPBACK}][/a]
BJL,
Why is there no contest involving double the photon count? There should be. As bjanes has written (and I agree with him on this point), the best tonality will be achieved by exposing fully to the right using the lowest ISO setting which permits an adequate shutter speed at the desired aperture. If I can double the photon count at a lower ISO without blowing wanted highlight detail, why should I not take that opportunity? What possible advantage is there in using a higher ISO than is needed?
Now, it makes some sense if you are advocating this technique (of achieving ETTR through increased preamplification rather than increased photon count) as a personal style of shooting because (let's say) you have a tendency to underestimate the shutter speed required for a sharp image. This would be analagous to someone setting all clocks forward by a few minutes because he/she has a tendency to be late. (Not that that really works because one tends to be constantly aware that the clock is really 10 minutes fast ).
-
The above works well for film with exposure determined from middle gray. However, with ETTR in digital, one should place the highlight just short of clipping (data value 4095 in the raw file with a 12 bit AD converter). The easiest way to do this is probably to take a spot reading from the highlight
... which may involve purchasing another camera.
So it's not quite as easy as you 1-series zealots think.
(Yeah, yeah, I know the 30D has a "spot" metering mode, that the 5D really has it, and Nikonians are probably laughing right now, but I still think it was a funny point to raise.)
-
The above works well for film with exposure determined from middle gray. However, with ETTR in digital, one should place the highlight just short of clipping (data value 4095 in the raw file with a 12 bit AD converter). The easiest way to do this is probably to take a spot reading from the highlight and expose according to a previously determined offset from the nominal meter reading that places the highlights properly. This is often about 2 1/3 stops over the indicated exposure.
[a href=\"index.php?act=findpost&pid=63508\"][{POST_SNAPBACK}][/a]
How you describe "ETTR in digital" is exactly how I set the exposure for a slide in a film camera using manual mode and spot metering. If this is indeed the case, it would be a lot simpler for the ETTR writers to simply refer to this technique which many film camera users are already familiar with. Of course to use this technique, either the film or digital camera must support manual exposure, and preferrably has spot metering.
-
Here is an explanation of sensor ISO as it applies to digital sensors:
http://www.normankoren.com/digital_cameras.html#ISOspeed (http://www.normankoren.com/digital_cameras.html#ISOspeed)
[a href=\"index.php?act=findpost&pid=63003\"][{POST_SNAPBACK}][/a]
Thanks. Koren's site has a wealth of info.
-
How you describe "ETTR in digital" is exactly how I set the exposure for a slide in a film camera using manual mode and spot metering. If this is indeed the case, it would be a lot simpler for the ETTR writers to simply refer to this technique which many film camera users are already familiar with. Of course to use this technique, either the film or digital camera must support manual exposure, and preferrably has spot metering.
[a href=\"index.php?act=findpost&pid=63543\"][{POST_SNAPBACK}][/a]
Chris,
I love your answer, which is in accord with the maxim of expose for the highlights with transparency film. It may be nothing new to most of us, but it is apparently for the photo expert Jeff Schewe who maintains that exposure for digital is entirely different than with film.
The problem really boils down to one of convention...digital sensors are NOT like film. They do not react the way chrome nor neg film reacts to light. Therefore you really can't use old film exposure techniques...the aim is to get as much data captured as far up the scale as you can while still preserving usable textural highlight detail. This is NOT "over exposing" it's "proper exposing"-for a digital sensor.
[a href=\"index.php?act=findpost&pid=62533\"][{POST_SNAPBACK}][/a]
-
Ray, you still do not get what I am saying, even after I used italics in my previous post! There is no contest involving double the photon count. I am comparing different ISO speed setting options that involve exactly the same photon count at each photosite due to using exactly the same shutter speed and aperture, but different amounts of pre-amplification so that those same photon counts are converted to different voltages being input to the A/D converter, giving different digital levels going into the RAW output. My goal is "exposing the A/D convertor to the right", giving it the highest input voltage that it can handle so that the brightest parts of the image are represented in the RAW file by near-maximum digital levels.
And this is specifically about situations calling for higher than base ISO speed in order to get "on meter" exposure, so it is best to stay away from examples involving ISO 100!
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63486\")
I did an analysis of the following exposure situations according to the methods used by Roger Clark for the Canon EOS 1D Mark II. Interested readers should refer to Roger's web site for details.
[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]
1. Scene exposed for ISO 100 with camera set to ISO 100
2. Scene exposed for ISO 1600 with camera set to ISO 1600
3. Scene exposed for ISO 1600 with camera set to ISO 100
The results are shown in the table below. [attachment=506:attachment]
At ISO 100 the sensor wells contain 53,000 electrons at highlight exposure to the right where the raw file data number is 4071 (as calculated from the gain of 13.02 electrons/data number). If one exposes for ISO 1600, since the sensor is linear, it will have about 3313 electrons and the raw file data number will be 4090 (calculated from the gain of 0.81 electrons/DN). Read noise is 16.6 electrons at ISO 100 and 3.90 electrons at ISO 1600.
Under these shooting conditions, almost all of the noise is accounted for by shot (electron sampling or Poisson noise) and read noise. Dark current is not significant for these exposure times. The table shows the noise calculations for a 10 zone scene. The shot noise predominates at high signal values and the read noise for low values. The total calculated noise is also shown along with the signal to noise ratio (SN) for each zone and the data numbers for a 12 bit linear file and for gamma 2.2.
For the ISO 1600 exposure with the camera set to ISO 100, AD underflow will occur in the shadows assuming 12 bit integer math, and the model predicts that the shadows will be clipped. To apply +4 stop exposure compensation to the ISO 100 exposure, it is necessary to multiply the raw data number by 16.
Now if we look at Ray's test shots we do see shadow clipping in the ISO 1600 exposure with the camera set at ISO 100 and considerably higher shadow noise due to the higher read noise. Shot noise is the same since the number of photons reaching the sensor is the same in both cases. The model goes a long way in explaining the observed results without invoking any additional NR.
-
BJL,
Why is there no contest involving double the photon count?
[a href=\"index.php?act=findpost&pid=63516\"][{POST_SNAPBACK}][/a]
Ray (and bjanes)
I mean that the alternatives that I am discussing do not involve one with double the photon count of the other. I am discussing different ways of processing the same photon counts, by applying different degrees of pre-amplification (different ISO settings) to them.
This is different from the ETTR scenario, which involves adjusting the "sensor exposure level" by changing the shutter speed (or f-stop) to vary the amount of light gathered by each photo-site.
-
For the best signal to noise, one should use the lowest ISO that is feasable for the shooting conditions.
[a href=\"index.php?act=findpost&pid=63508\"][{POST_SNAPBACK}][/a]
More specifically, one should use the lowest shutter speed that is feasible (assuming fixed f-stop), because that maximizes the among of light received by each photo-site, and so maximizes the S/N ratio of the signal coming out of the photo-sites (reflecting combined dark current and photo noise).
The amplifier gain is then increased to compensate for the reduced exposure, so that the highlights will be at the desired location in the histogram.[a href=\"index.php?act=findpost&pid=63508\"][{POST_SNAPBACK}][/a]
This is back-to-front. ETTR is about increasing exposure level to get as much light as possible on each photo-site, which goes with _reducing_ amplifier gain: lower ISO speed setting = lower amplifier gain.
But you are still talking about the scenario where decreasing ISO speed goes with decreasing shutter speed. I am discussing the different, high speed/low light situation where I am up against the minimum acceptable shutter speed, and might choose an ISO setting other that the "normal" one suggested by the light meter.
-
I did an analysis of the following exposure situations according to the methods used by Roger Clark for the Canon EOS 1D Mark II. Interested readers should refer to Roger's web site for details.
http://www.clarkvision.com/imagedetail/eva...-1d2/index.html (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html)
1. Scene exposed for ISO 100 with camera set to ISO 100
2. Scene exposed for ISO 1600 with camera set to ISO 1600
3. Scene exposed for ISO 1600 with camera set to ISO 100
[a href=\"index.php?act=findpost&pid=63558\"][{POST_SNAPBACK}][/a]
Thanks, I am talking about a milder case of the comparison between the last two cases: same exposure, different ISO setting (In my case, maybe exposing for ISO 800 but with camera set to ISO 800 vs cameras set to ISO 1600, in a situation when the later will not ruin highlights.)
As the table show, setting the camera to a higher ISO (with exposure levels based on the same ISO) improves the S/N ratios by reducing the effect of read noise.
I am not sure about dark current noise being negligible though; for Kodak's FF CCD DSLR sensors, dark current noise can contribute about 16-22 electrons of noise, comparable to or greater than the read noise figures that Clark gives. AFAIK, CCD dark current noise is roughly constant for all but very long exposures (1s or longer) because it continues to accumulate after the exposure is finished but the electrons are still waiting to be read out.
Maybe though, Canon's CMOS sensor do immediate pre-amplifyication "on site" while waiting for read out, avoiding this extra accumulation of dark current noise. That might be part of their good shadow noise handling.
-
Thanks, I am talking about a milder case of the comparison between the last two cases: same exposure, different ISO setting (In my case, maybe exposing for ISO 800 but with camera set to ISO 800 vs cameras set to ISO 1600, in a situation when the later will not ruin highlights.)
As the table show, setting the camera to a higher ISO (with exposure levels based on the same ISO) improves the S/N ratios by reducing the effect of read noise.
I am not sure about dark current noise being negligible though; for Kodak's FF CCD DSLR sensors, dark current noise can contribute about 16-22 electrons of noise, comparable to or greater than the read noise figures that Clark gives. AFAIK, CCD dark current noise is roughly constant for all but very long exposures (1s or longer) because it continues to accumulate after the exposure is finished but the electrons are still waiting to be read out.
Maybe though, Canon's CMOS sensor do immediate pre-amplifyication "on site" while waiting for read out, avoiding this extra accumulation of dark current noise. That might be part of their good shadow noise handling.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63588\")
According to Roger's tests, dark current averages for the EOS 1D Mark II at ISO 1600 were 0.013 to 0.02 electrons/second, but some pixels have dark currents as high as about 0.25 electrons/second. For exposures of a few seconds, this noise is neglibible, but for exposures in minutes, the dark current then would be significant. The camera most likely has a dark current subtraction option, at least the Nikon's do, but it is not needed even with the Nikon (which has poorer noise characteristics) under normal circumstances. See Roger's site for details. Also, these methods could easily be adapted to your situation.
Increasing the ISO may reduce read noise, but that is a relative minor component of noise in most situations it is outweighed by photon noise as shown in the spreadsheets.
Your point about the dark noise accululation post exposure is a good one; Roger waited for the data to be written out to memory from the buffer before taking more shots, so the delay should be minimal.
[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]
-
More specifically, one should use the lowest shutter speed that is feasible (assuming fixed f-stop), because that maximizes the among of light received by each photo-site, and so maximizes the S/N ratio of the signal coming out of the photo-sites (reflecting combined dark current and photo noise).
This is back-to-front. ETTR is about increasing exposure level to get as much light as possible on each photo-site, which goes with _reducing_ amplifier gain: lower ISO speed setting = lower amplifier gain.
But you are still talking about the scenario where decreasing ISO speed goes with decreasing shutter speed. I am discussing the different, high speed/low light situation where I am up against the minimum acceptable shutter speed, and might choose an ISO setting other that the "normal" one suggested by the light meter.
[a href=\"index.php?act=findpost&pid=63585\"][{POST_SNAPBACK}][/a]
I don't use your shooting methods and have not read your posts in detail. However, as I said, for the best S/N one should use the lowest ISO possible, since that will maximize the photon count--this is entirely different from exposing to the right as discussed by Mr. Reichman and others. Exposing fully to the right at ISO 1600 will give a more noisy image than 1 stop underesposure at ISO 100 as shown on the spreadsheet that I posted previously. Of course, for best results, one should expose to the right with ISOs of 100 or 1600.
Of course amplifier gain is less at low ISO and I don't know why you even bother to mention this. ETTR per se, does not involve choice of an ISO, but merely adjusting the exposure so that the AD converter output is near maximum (DN 4095 for a 12 bit device). This can occur at ISO 100 or 1600.
I don't know about your light meter, but with meters I use one enters the ISO and the meter then indicates the exposure. One could work backwards, but I normally do not do so.
-
More specifically, one should use the lowest shutter speed that is feasible (assuming fixed f-stop), because that maximizes the among of light received by each photo-site, and so maximizes the S/N ratio of the signal coming out of the photo-sites (reflecting combined dark current and photo noise).
[a href=\"index.php?act=findpost&pid=63585\"][{POST_SNAPBACK}][/a]
BJL,
As I suspected, you are really suggesting a procedural approach for achieving the optimum combination of shutter speed, aperture and ISO, consistent with ETTR.
I guess we all have a slightly different approach to getting the most out of our camera, and that approach will also be influenced by the design of the camera.
The 2 cameras I've used most to date are the D60 and the 5D. The 20D was an interim purchase which I would never have made if I'd known at the time that the 5D was in development.
A frustrating factor in using the D60 resulted form the fact that ISO could not be changed whilst looking through the viewfinder. One had to to press a menu button, scroll down to the ISO settings, press another button, turn a wheel, then press a button again. If you are long sighted, as I am, then that would also involve putting on one's reading glasses. The whole process was a major interruption to the act of taking a photo. I therefore did a series of tests to determine just how much benefit to image quality and tonality was achieved by using a higher ISO setting, as opposed to using the same exposure at ISO 100 and in effect underexposing the image.
To my surprise, there was essentially none, or very little at best. The deepest shadows showed marginally better detail and tonality, but such shadows are usually black on a print. In general, not only were the differences marginal but they were only apparent at a pixel-peeping level at 200% enlargement on screen, which is equivalent to a far bigger print than I would ever make from a 6MP image.
The consequence of this is that I didn't use the higher ISO settings on the D60 as much as I should have. I tended to rely upon IS to reduce camera shake and too often forgot that a shutter speed that is adequate for camera shake is not necessarily adequate for even a slowly moving subject.
Old habits die hard and this is perhaps one reason why I'm so relaxed about underexposing a shot at ISO 100. One gets the benefit of a faster shutter speed and the greater certainty of retaining all highlight detail. Essentially, the D60 was a one ISO camera with different settings for brightening the LCD review. If you couldn't get a fast enough shutter speed at ISO 100, then you were basically stuffed. There was no alternative to a noisy image, outside use of a tripod.
The 5D however is a very different camera. As far as I can tell, image quality at ISO 100 is not one whit better, appart from the greater resolution afforded by more pixels. My first impression was that it is actually worse, in terms of S/N and DR, than the D60. This is because the 5D suffers from noticeable banding which the D60 is free of. Underexpose a high DR scene by one stop with the 5D and objectional banding is very apparent when the shadows are lightened. Underexpose the same scene by one stop with the D60 and the shadows are far less objectionable (but still degraded of course in terms of detail and tonality).
However, I need to do more tests to confirm if this. There's a slight discrepancy between the true ISO values for both cameras at ISO 100. On the 5D, ISO 100 is actually 125 and on the D60 I think it's around 150 or 160.
-
I did an analysis of the following exposure situations according to the methods used by Roger Clark for the Canon EOS 1D Mark II. Interested readers should refer to Roger's web site for details.
[a href=\"index.php?act=findpost&pid=63558\"][{POST_SNAPBACK}][/a]
bjanes,
It's taken me a while to wrap my head around these results you've shown in the table. I'm not specifically technically orientated. As a layman, I sometimes struggle to find a meaningful interpretation from tables of figures, so perhaps you can help me out here.
Let's do an analysis of a single row (zone 7) taken from the exposure situation equivalent to my previous test images in the thread.
I've reproduced the two lines of data for easy reference.
[attachment=507:attachment]
1. From left to right the total number of electrons and shot noise is the same, 414 and 20.3 (the square root of 414).
2. The read noise is significantly higher at ISO 100, 16.6 as opposed to 3.9, yet the total noise does not reflect a simple addition of both noise figures. At ISO 100, the total noise is only 5.6 electrons greater than at ISO 1600. 5.6 electrons greater noise in relation to 414 does not seem significant.
3. The S/N of 15.8 as compared with 20 for the ISO 1600 shot might seems significant, but this figure is a ratio and might be misleading.
4. The huge discrepancy I see, comparing these figures, is in the actual quantity of data numbers (DN) available to describe or quantify the signal during A/D conversion.
Whether we've set the camera to ISO 100 or 1600, the possible total variation in electron count in zone 7 (roughly mid-tones) is 414. My simple mind tells me that within that range of exposure (roughly one f/stop, but probably a bit less), there are 414 possible values, but only 31 numbers to describe those values at ISO 100. At ISO 1600, however, there are 511 numbers to describe the 414 electrons.
This I would suggest is the major reason for the dramatic reduction in noise at ISO 1600.
However, having said that, there is the problem of the D60 showing only marginal noise improvement in the same circumstances. We can therefore assume that the D60 employes 'noisy' preamplifiers, whereas the 5D and the 1D2 employ less noisy preamplifiers. Alternatively, we could surmise that it's a combination of superior components plus a better arrangement of processes on the chip.
Either way, it becomes an argument in semantics to quibble about whether or not additional noise reduction has taken place. Noise in the D60 at ISO 100 is no worse than noise in the 5D at ISO 100. Noise in the D60 at ISO 1000 (the highest setting) is very much worse than noise in the 5D at ISO 1000.
-
bjanes,
It's taken me a while to wrap my head around these results you've shown in the table. I'm not specifically technically orientated. As a layman, I sometimes struggle to find a meaningful interpretation from tables of figures, so perhaps you can help me out here.
Let's do an analysis of a single row (zone 7) taken from the exposure situation equivalent to my previous test images in the thread.
I've reproduced the two lines of data for easy reference.
[attachment=507:attachment]
1. From left to right the total number of electrons and shot noise is the same, 414 and 20.3 (the square root of 414).
2. The read noise is significantly higher at ISO 100, 16.6 as opposed to 3.9, yet the total noise does not reflect a simple addition of both noise figures. At ISO 100, the total noise is only 5.6 electrons greater than at ISO 1600. 5.6 electrons greater noise in relation to 414 does not seem significant.
3. The S/N of 15.8 as compared with 20 for the ISO 1600 shot might seems significant, but this figure is a ratio and might be misleading.
4. The huge discrepancy I see, comparing these figures, is in the actual quantity of data numbers (DN) available to describe or quantify the signal during A/D conversion.
Whether we've set the camera to ISO 100 or 1600, the possible total variation in electron count in zone 7 (roughly mid-tones) is 414. My simple mind tells me that within that range of exposure (roughly one f/stop, but probably a bit less), there are 414 possible values, but only 31 numbers to describe those values at ISO 100. At ISO 1600, however, there are 511 numbers to describe the 414 electrons.
This I would suggest is the major reason for the dramatic reduction in noise at ISO 1600.
However, having said that, there is the problem of the D60 showing only marginal noise improvement in the same circumstances. We can therefore assume that the D60 employes 'noisy' preamplifiers, whereas the 5D and the 1D2 employ less noisy preamplifiers. Alternatively, we could surmise that it's a combination of superior components plus a better arrangement of processes on the chip.
Either way, it becomes an argument in semantics to quibble about whether or not additional noise reduction has taken place. Noise in the D60 at ISO 100 is no worse than noise in the 5D at ISO 100. Noise in the D60 at ISO 1000 (the highest setting) is very much worse than noise in the 5D at ISO 1000.
[a href=\"index.php?act=findpost&pid=63621\"][{POST_SNAPBACK}][/a]
Ray,
1. The noises are expressed in standard deviations and you may remember from statistics that SDs do not add, but variances do. So the total noise is the square root of the sum of the two standard deviations squared.
3. The S/N is the standard for evaluation of noise. If you look at the ISO 100 exposures, you will see that they have the highest noise levels, but since the S/N is high, noise is not visible.
4. I did look at the histogram of the tests for you repeat test showing the trunk of a tree. The ISO 100 shot shows clipping of the shadows as predicted by the spreadsheet and the shadows are replaced by noise. The noise does look higher than predicted by the model, but I don't really know how the 5D is responding here or if there is additional noise reduction at ISO 1600. However, at ISO 100 with a 12 bit data number of 254, you are making use of only the lower 8 bits of the AD converter and are thorwing away resolution. With ISO of 1600, the DN is 4090 and you are making use of the full range of the AD converter.
To me, it does not make sense to set the camera for ISO 100 and then expose as for ISO1600.
-
bjanes,
I've reproduced the two lines of data for easy reference.
[attachment=507:attachment]
4. The huge discrepancy I see, comparing these figures, is in the actual quantity of data numbers (DN) available to describe or quantify the signal during A/D conversion.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63621\")
I did not answer your final quesiton. Conversion from numbers of electrons to camera data number involves the camera gain, which is expressed in electrons/12 bit data number. For ISO 100, the gain of the EOS 1D MII is 13.02 electrons/DN and for ISO 1600 the gain is 0.81 electrons/DN. Gain reported in CCD imaging is actually inverse amplification; that is why it decreases with increasing ISO.
[a href=\"http://www.photomet.com/library_enc_gain.shtml]http://www.photomet.com/library_enc_gain.shtml[/url]
Spending a bit of time reading Roger's post and his references is time very well spent.
-
BJL,
As I suspected, you are really suggesting a procedural approach for achieving the optimum combination of shutter speed, aperture and ISO, consistent with ETTR.
That sounds right: I am trying to adopt the digital reality of being able to easily use all three exposure parameters (aperture, shutter speed and "gain/sensitivity" (ISO speed)).
I therefore did a series of tests [with a D60] to determine just how much benefit to image quality and tonality was achieved by using a higher ISO setting, as opposed to using the same exposure at ISO 100 and in effect underexposing the image.
To my surprise, there was essentially none, or very little at best. The deepest shadows showed marginally better detail and tonality, but such shadows are usually black on a print.
That makes sense to me; common or garden variety JPEG's from DSLR's at low to moderate ISO settings typically have about an eight stop DR, a couple of stops more than prints can reveal. So this ETTR and ISO setting fiddling is mainly relevant when dealing with a scene of high subject brightness range (over 8 stops), so that one needs to compress that range for printing, including bringing deep shadows up several stops.
-
For ISO 100, the gain of the EOS 1D MII is 13.02 electrons/DN and for ISO 1600 the gain is 0.81 electrons/DN. Gain reported in CCD imaging is actually inverse amplification; that is why it decreases with increasing ISO.
I'm not implying that there is an error in these figures. I'm stating that as a consequence of preamplification there are more data numbers available to describe each level (voltage variance, however you want to describe it) for each photosite. It is clearly better to have more than one DN for each electron than less than one DN. One DN for 13 electrons seems clearly inadequate and seems more likely to be the cause of the greater noise in the ISO 100 image than increased read noise.
The problem I have with your assertion (admittedly because you believe more qualified people have made that assertion) that there is no additional noise reduction taking place at higher ISOs and that the greater noise readily apparent in the ISO 100 image is due primarily to additional read noise, is that it raises the question, 'why have Canon not implemented 16 bit A/D conversion?'
A 16 bit A/D converter for a fully exposed image might well be overkill, but it would certainly seem that it could help the underexposed image (or underexposed part of the image) at ISO 100.
It would seem to me, if you are right, that all Canon has to do to dramatically improve dynamic range is reduce readout noise at ISO 100 to the same level it is at ISO 1600 and then use 16 bit A/D conversion. Since the nature of the CMOS chip is that it can accommodate additional processors at the photosite, that does not seem, on the face of it, an insurmountable problem.
If you are right (and actually I'd prefer you to be right) we can look forward to a future DSLR which will be as noise-free in the deepest shadows at ISO 100 as the 1D2 and 5D currently are at ISO 1600 (with the same photon count).
bjanes, I read the explanation of gain in the link you provided. I've read similar stuff years ago. There's nothing new there for me. What is new is the concept that it's possible for a camera system to produce images that are dramatically less noisy at a given ISO setting and similar pixel size (or pitch) without having employed additional noise reduction.
There is clearly something different between the preamplification of a signal in the earlier D60 and the preamplification of the same signal in the 20D. If it doesn't involve noise reduction, or noise cancellation of some sort, then what does it involve?
-
That sounds right: I am trying to adopt the digital reality of being able to easily use all three exposure parameters (aperture, shutter speed and "gain/sensitivity" (ISO speed)).
[a href=\"index.php?act=findpost&pid=63685\"][{POST_SNAPBACK}][/a]
Now that cameras like the 20D and 5D show such a marked improvement at higher than base ISO, underexposure at ISO 100 suddenly seems a disadvantage in relative terms. With the D60 is wasn't.
I am now wondering if the next step in a future Canon DSLR will be another 'priority' mode. In addition to AV and TV priority, I would find advantageous a combined AV & TV priority mode in relation to ISO. In other words, the user selects the aperture and shutter speed, as in manual mode, and the camera automatically selects the approprate ISO for the metered reading. It would save time. Getting the aperture and shutter speed right for a particular shot is critical for a sharp image with a desired DoF. In situations where one is trying to capture the moment, AV can let one down by producing a shutter speed that's too slow. TV can result in a DoF which is too shallow, and full manual mode can cause one to waste time trying to select an approprite ISO, which could mean missing the shot.
-
I'm not implying that there is an error in these figures. I'm stating that as a consequence of preamplification there are more data numbers available to describe each level (voltage variance, however you want to describe it) for each photosite. It is clearly better to have more than one DN for each electron than less than one DN. One DN for 13 electrons seems clearly inadequate and seems more likely to be the cause of the greater noise in the ISO 100 image than increased read noise.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63689\")
It really does not make sense to have the camera gain set to less than less than the readout noise. For an explanation see the link below. For example, the the EOS 1D Mark II at ISO 100 has a gain of 13 electrons/12 bit DN and a read noise of 16 electrons. If you had a 16 bit AD converter, you could set the gain to 1 electron/DN and not have overflow at the full well electron count, but you would be resolving mainly noise.
[a href=\"http://spiff.rit.edu/classes/phys559/lectures/gain/gain.html]http://spiff.rit.edu/classes/phys559/lectures/gain/gain.html[/url]
The problem I have with your assertion (admittedly because you believe more qualified people have made that assertion) that there is no additional noise reduction taking place at higher ISOs and that the greater noise readily apparent in the ISO 100 image is due primarily to additional read noise, is that it raises the question, 'why have Canon not implemented 16 bit A/D conversion?'
[a href=\"index.php?act=findpost&pid=63689\"][{POST_SNAPBACK}][/a]
Probably for the above reason. Why increase the raw file size merely to record noise? Ron Parr and others have not denied that the Canon employs additional noise reduction at ISO 1600, but merely that they see no evidence of it. Why do not bring your observations to Ron's attention? He is a regular visitor on the DPReview forums. Since he shoots Canon, you might try the Canon forums.
bjanes, I read the explanation of gain in the link you provided. I've read similar stuff years ago. There's nothing new there for me. What is new is the concept that it's possible for a camera system to produce images that are dramatically less noisy at a given ISO setting and similar pixel size (or pitch) without having employed additional noise reduction.
[a href=\"index.php?act=findpost&pid=63689\"][{POST_SNAPBACK}][/a]
Well, Ray, if you do not know that standard deviations are not additive and that it makes little sense to have the camera gain set to less than the readout noise, then you probably should do a bit more reading.
-
Well, Ray, if you do not know that standard deviations are not additive and that it makes little sense to have the camera gain set to less than the readout noise, then you probably should do a bit more reading.
[a href=\"index.php?act=findpost&pid=63712\"][{POST_SNAPBACK}][/a]
bjanes,
There are not enough hours in the day to do the reading required to know all the things that even I think I should know, never mind what you think I should know.
I read what interests me and I'm fast losing interest in this topic. Let's call it quits.
-
I've been following this thread with my mouth shut (i.e. fingers off the keyboard) because I agree with expose to the right, but I'm in learning mode about the relationship between ISO settings and visible noise. I was always told and read, until I started seeing the material on this thread, that lower ISO settings produce less noise than higher ones - in general. So in just about every camera review one reads, one is told the camera is fine at ISO 100, and then depending on the model they get increasingly noisy as you progress up beyond say 400, etc. Those observations seem contrary to views in this thread that shooting at high ISO produces LESS visible noise than shooting at low ISO (and I assume that whether we use low or high ISO we are keeping the histogram to the right without clipping, hence shutter speed/and or aperture adjusts accordingly). Or have I not understood correctly what is being said here? Can someone explain in simple English why shooting at high ISO produces less visible noise than shooting at low ISO, assuming one obeys the histogram to the right principle, and of course one is using the same camera?
Ray - not time to quit yet. Let's see if "civil clarity" emerges on this point.
-
So in just about every camera review one reads, one is told the camera is fine at ISO 100, and then depending on the model they get increasingly noisy as you progress up beyond say 400, etc. Those observations seem contrary to views in this thread that shooting at high ISO produces LESS visible noise than shooting at low ISO (and I assume that whether we use low or high ISO we are keeping the histogram to the right without clipping, hence shutter speed/and or aperture adjusts accordingly).
Mark,
Sorry for the confusion here. We've introduced a comparison which is not normally made; an exposure that is correct for ETTR at ISO 1600 compared with the same exposure at ISO 100, or to put it another way a fully exposed shot at ISO 1600 compared with a shot at ISO 100 that is underexposed by 4 stops.
The purpose of doing this seems of more academic interest than of practical value, but the results are very surprising for me. They really put into graphic terms just how much cleaner high ISO image now are in Canon's latest DSLRs, yet images at base ISO do not seem to have got any better, whether they are underexposed or exposed to the right. A 5D image at ISO 100 is just as noisy in the shadows as my old D60.
The conclusion I would draw from this is that at base ISO (100), it is now even more important to expose to the right because there is more to lose by not doing so. That's is, there's a higher ISO alternative which really does reduce noise at the same exposure.
Has this clarified anything?
ps. I mentioned earlier in the thread that I felt quite relaxed about not exposing to the right at ISO 100 and endangering highlight detail loss. Having thought this through, I'm now not so relaxed. With my D60, a stop underexposed at ISO 100 was really no different to a fully exposed shot at ISO 200. If I needed a particular shutter speed and aperture that resulted in a stop of underexposure at ISO 100, then so be it. Of course I'm talking about RAW images and hand-held shots and situations where it's inconvenient to change ISO.
-
Yes and no. It raises a further question about whether these comparisons are of analytic value - by which I mean are we isolating one variable at a time and seeing the effect. From what I've been reading, it seems that one should expect less noise with lower ISO, and less noise the further to the right the image sits on the histogram. If that is the case, when you crank the ISO to 1600 you introduce noise, but then when you crank down the ISO to 100, you introduce noise because of under-exposure. So then the issue becomes whether under-exposure at a "clean" ISO produces more or less noise than correct exposure at a "less clean" ISO. May-be of some interest to some shooting situations, but I guess I was thinking more in terms of "best practice" in general. It still seems to be, after all this discussion - that best practice consists of using the lowest ISO achievable with ETTR conditions, where "achievable" means a shutter speed and aperture that makes photographic sense relative to the subject matter. Is this correct?
-
It still seems to be, after all this discussion - that best practice consists of using the lowest ISO achievable with ETTR conditions, where "achievable" means a shutter speed and aperture that makes photographic sense relative to the subject matter. Is this correct?
[a href=\"index.php?act=findpost&pid=63748\"][{POST_SNAPBACK}][/a]
Basically, yes, with a qualification that I added in the postcript in my previous thread. But I'm struggling to understand why Canon are not improving noise at base ISO yet are achieving such spectacular results at high ISO.
-
Yes and no. It raises a further question about whether these comparisons are of analytic value - by which I mean are we isolating one variable at a time and seeing the effect. From what I've been reading, it seems that one should expect less noise with lower ISO, and less noise the further to the right the image sits on the histogram. If that is the case, when you crank the ISO to 1600 you introduce noise, but then when you crank down the ISO to 100, you introduce noise because of under-exposure. So then the issue becomes whether under-exposure at a "clean" ISO produces more or less noise than correct exposure at a "less clean" ISO. May-be of some interest to some shooting situations, but I guess I was thinking more in terms of "best practice" in general. It still seems to be, after all this discussion - that best practice consists of using the lowest ISO achievable with ETTR conditions, where "achievable" means a shutter speed and aperture that makes photographic sense relative to the subject matter. Is this correct?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63748\")
Mark,
I think your summary is right on.
You might want to look at another thread on the Nikonians forum by Bill Claff, who seems quite knowledgable. He finds that for the Nikon D70, noise is the same for ISO 100 and ISO 1600, provided that the exposure is the same and the same number of electrons are collected by the sensor:
[a href=\"http://home.comcast.net/~TomatoMan/Mirror/Cached/86_4938.htm]http://home.comcast.net/~TomatoMan/Mirror/Cached/86_4938.htm[/url]
I did a quick test with my Nikon D200, exposing a scene at ISO 100 and 1600, using the same shuttter speed and aperture. My results more or less confirm Bill's and I don't see much difference in noise. The ISO 1600 is mislabeled as ISO 16000. The NEFs were processed in ACR with exposure +4EV for the ISO 100 shot. These results are unlike Ray's and require explanation--I don't have the answer. Perhaps Canon is doing some NR.
Also here is a detailed noise analysis of the D200 at ISOs 100 and 1600 according to Roger Clark's methods. As is shown, the ISO 100 images have much better S/N. I see no advantage exposing with unorthodox methods.
-
best practice consists of using the lowest ISO achievable with ETTR conditions, where "achievable" means a shutter speed and aperture that makes photographic sense relative to the subject matter. Is this correct?
[a href=\"index.php?act=findpost&pid=63748\"][{POST_SNAPBACK}][/a]
This discussion might be simplified of we put our imaginary cameras in full manual exposure setting mode, instead of the aperture priority mode that seems to be silently assumed in many comparisons. That is, we should not assume that shutter speed and ISO speed are tied together.
Then the first key to minimizing noise is using the lowest possible shutter speed. If you are using aperture priority mode with no exposure compensation, this is the same as using the lowest possible ISO speed, but it is slightly misleading to attribute the low noise to the low ISO; the basic physical origin of the low noise (high S/N ratio) is the fact that a lower shutter speed gathers more light from the subject (at a given aperture.)
However, at a given shutter speed (and aperture), manual exposure setting allows more than one choice of ISO speed. For example we could choose between
(-2) f/2.8 1/250s ISO 100, with the meter saying we are two stops underexposed, probably requiring compensating up by two stops in post-processing.
(-1) f/2.8 1/250s ISO 200, with the meter saying one stop underexposed
(0) f/2.8 1/250s ISO 400, with the meter saying correct exposure
(+1) f/2.8 1/250s ISO 800, with the meter saying one stop overexposed
(+2) f/2.8 1/250s ISO 1600, with the meter saying two stops overexposed
Now, increasing ISO here will in general improve the S/N ratio, by reducing the impact of read noise and quantization noise in the A/D converter. I think that we are all agreed that option (-2) is probably a bad idea, (-1) is likely to be better, (0) to be better still.
But why stop at (0)? Why not keep going to (+1) or (+2) or so on?
Answer: at some point, the extra pre-amplification of the higher ISO speeds will cause highlight values to exceed the limits of the amplifier ("clipping"), or to exceed the input level that the A/D convertor can handle: let me call either of these problems "clipping".
But if you stop short of clipping, so that the highlights are pre-amplified to about the maximum level that the pre-amp can handle and the maximum input level of the A/D converter, producing near maximum output levels (about 4096 with 12-bit A/D), the results are probably optimal. This is likely to be far better than with one fourth the ISO speed, when the signal to the A/D converter is two stops lower so that the brightest highlights come out at only about level 1024 out of 4096 in 12-bit A/D output.
Most often, option (0) will be best, but for a scene no highlights much brighter than the metered mid-tones and with interesting detail in deep shadows, option (+1) might be better. And if the metered is dominated by a subject at low reflectivity (say 10%) and there are a few highlights at close to 100% reflectivity, over three stops brighter than the metered mid-tones, option (0) might cause highlight clipping, so that "under-amplification" like option (-1) might be better.
In other words, choose
a shutter speed and aperture that makes photographic sense relative to the subject matter
[a href=\"index.php?act=findpost&pid=63748\"][{POST_SNAPBACK}][/a]
and then decide whether to use the normal "matching" ISO speed or a different one.
Ray has the same idea as I have speculated about: an "ATv" auto exposure mode where one specifies shutter speed, aperture and maybe exposure compensation, from which the AE system chooses the ISO speed. Better yet, if the light metering could effectively scan all photo-sites to find the brightest highlights, it could choose an ISO speed just from shutter speed and aperture n order to have the brightest highlights come out at "level 4096".
P. S. But as Ray and I seem to agree, probably only in rare cases do deep shadows need this much care and feeding.
-
Most often, option (0) will be best, but for a scene no highlights much brighter than the metered mid-tones and with interesting detail in deep shadows, option (+1) might be better. And if the metered is dominated by a subject at low reflectivity (say 10%) and there are a few highlights at close to 100% reflectivity, over three stops brighter than the metered mid-tones, option (0) might cause highlight clipping, so that "under-amplification" like option (-1) might be better.
[a href=\"index.php?act=findpost&pid=63788\"][{POST_SNAPBACK}][/a]
Look at Jeff Schewe's comments earlier in this thread. If you are metering from the midtones, you really are not exposing properly for digital--you should be metering and exposing for the highlights. Until such time as the camera system can scan all the photosites and find the brightest highlights as you mention, you will have to do this yourself. Also, the automated system might not be able to exclude specular highlights. With high dynamic range subjects, it would not know whether to preserve highlights or shadows.
The differences in read noise at various ISOs are relatively insiginificant in most situations (as shown by Dr. Clark's analysis), but are of interest to astronomers. If you really need good SN in the shadows, you should really strive to expose as near to full well capacity of the sensor as possible, and expose maximally to the right. A half stop or so of blown highlights can usually be recovered in ACR.
-
I did a quick test with my Nikon D200, exposing a scene at ISO 100 and 1600, using the same shuttter speed and aperture. My results more or less confirm Bill's and I don't see much difference in noise. [a href=\"index.php?act=findpost&pid=63754\"][{POST_SNAPBACK}][/a]
I think you might have just opened a can of worms with this latest comparison, bjanes. I'm surprised BJL has not commented.
Since I don't own any Nikon equipment, I tend not to spend much time reading Nikon reviews, but there's a very clear demonstration at dpreview of the effect of turning on and off the in-camera noise reduction of the D200, at high ISOs. I gather the D200's NR feature is very analagous to the software NR of programs like Neat Image, ie. there's usually a trade-off with regard to fine detail.
I don't know whether or not some users prefer to leave the camera's NR off and do their own noise reduction with Noise Ninja or Neat Image, but it wouldn't surprise me.
One final comment, and perhaps just another way of making the same point that BJL has already made, the ETTR principle is not always of much practical value in certain situations when one is battling to achieve a minimum shutter speed, depending upon the design of the camera.
I'll give a specific example. The ISO setting of 3200 on the 5D, by all accounts, does not provide better image quality than the same exposure would at ISO 1600 (which would be one stop underexposed by the meter).
The noise levels are generally so good on the 5D at high ISOs, one can wander around at night in any city and take candid shots without an obtrusive flash that draws attention to oneself. (In any case, the 5D does not sport a built-in flash).
When doing this, and I've done it, there's a tendency to set the camera on ISO 3200 (actually 'H') and be careful one is exposing fully to the right. One makes a judgement as to what minimum shutter speed would be acceptable. With the Canon 24-105 IS, I consider 1/25th as a minimum speed. 1/13th is too risky.
However, accepting that ISO 3200 provides no benefit over ISO 1600, noise-wise, the above procedure or practice does not seem the best. In that situation described above, not exposing to the right at a lower ISO is a better procedure.
There actually are advantages in not exposing to the right.
What are they? If the 5D is set on ISO 1600, the shutter speed 1/25th and the aperture maximum, yet exposure is 1 stop under, one should take the shot. Moving up to ISO 3200 to achieve ETTR will not improve image quality, will not improve the noise situation in the shadows, but will risk blowing out highlight detail in street lights and objects close to the lights. There will also be a risk of 'missing the moment' because it does take time to change ISO. There will also be an additional risk of missing the moment in the next shot, should lighting conditions improve slightly so that ETTR can be achieved by moving back to ISO 1600.
-
That makes sense to me; common or garden variety JPEG's from DSLR's at low to moderate ISO settings typically have about an eight stop DR, a couple of stops more than prints can reveal. So this ETTR and ISO setting fiddling is mainly relevant when dealing with a scene of high subject brightness range (over 8 stops), so that one needs to compress that range for printing, including bringing deep shadows up several stops.
[a href=\"index.php?act=findpost&pid=63685\"][{POST_SNAPBACK}][/a]
Are you suggesting that DSLRs can capture a DR range equal to negative film and greater than transparencies? Is this specific for DSLRs only, or all digital cameras? My understanding is that digital cameras' DR range (before any editing in RAW or PS) is similar to transparencies', i.e. about five stops.
Print compression occurs on both film and digital captures. But with PS (and RAW), some tricks can be applied to alleviate the print compression.
-
If you are metering from the midtones, you really are not exposing properly for digital--you should be metering and exposing for the highlights.
[a href=\"index.php?act=findpost&pid=63803\"][{POST_SNAPBACK}][/a]
Metering from the midtones (or auto metering modes) can be problematic for some images, on *both* film and digital cameras. I think that those who have mastered proper metering (actaully all aspects of photo techniques) on film have some advantage when crossing over to digital. It is quite amusing to find topics like composition, etc. being extensively covered in *digital* photography books, as if these are *digital* specific.
-
I think you might have just opened a can of worms with this latest comparison, bjanes. I'm surprised BJL has not commented.
Since I don't own any Nikon equipment, I tend not to spend much time reading Nikon reviews, but there's a very clear demonstration at dpreview of the effect of turning on and off the in-camera noise reduction of the D200, at high ISOs. I gather the D200's NR feature is very analagous to the software NR of programs like Neat Image, ie. there's usually a trade-off with regard to fine detail.
I don't know whether or not some users prefer to leave the camera's NR off and do their own noise reduction with Noise Ninja or Neat Image, but it wouldn't surprise me.
[a href=\"index.php?act=findpost&pid=63812\"][{POST_SNAPBACK}][/a]
Ray,
Yes, one has to be quite careful when reading noise tests on the D200 (and other Nikon cameras) since these are often done with JPEGs, where the in camera NR is automatically applied at ISO above 800 and can't be turned off. Noise may look good, but detail suffers, since as you state, the Nikon NR is a filtering process applied across pixels. Also, many cameras reduce sharpening with high ISO so as to avoid accentuating the noise. However, ACR ignores NR and sharpening settings that are tagged to the RAW file, and no NR is performed except for what is specified in the ACR detail tab. I usually use the default of 0 for luminance and 25 for color and no ACR sharpening. I then use Noise Ninja in PP.
The D200 does have quite a bit more read noise than the EOS 1DS Mark II--9.87 electrons vs 3.90 for the Canon for ISO 1600, so the methods you and BJL suggest for the Canon should work even better for the Nikons. However, Bill Claff and I have observed little difference.
Ron Parr has said that the Canon on chip NR at high ISO is consistent with the correlated double sampling that he thinks Canon does on chip and does not affect detail since it is not applied across pixels and is more like true NR and not simply filtering. In any event, methods to reduce shadow noise need to take the characteristics of the camera into account.
-
Are you suggesting that DSLRs can capture a DR range equal to negative film and greater than transparencies? Is this specific for DSLRs only, or all digital cameras? My understanding is that digital cameras' DR range (before any editing in RAW or PS) is similar to transparencies', i.e. about five stops.
Print compression occurs on both film and digital captures. But with PS (and RAW), some tricks can be applied to alleviate the print compression.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63825\")
See here (first table) as an example for the 30D:
[a href=\"http://www.dpreview.com/reviews/canoneos30d/page21.asp]http://www.dpreview.com/reviews/canoneos30d/page21.asp[/url]
According to Phil's measurements, the usable DR is about 8.4 EV until ISO 1600, where it drops to about 8 EV. At 3200 it's about a stop less, 7.3 EV. If you'd like more specifics, take a closer look at Phil's description of how he measures DR.
These measurements are for JPEGs. The bottom tables on that page are for RAW.
Eric
-
Are you suggesting that DSLRs can capture a DR range equal to negative film and greater than transparencies? Is this specific for DSLRs only, or all digital cameras? My understanding is that digital cameras' DR range (before any editing in RAW or PS) is similar to transparencies', i.e. about five stops.
Print compression occurs on both film and digital captures. But with PS (and RAW), some tricks can be applied to alleviate the print compression.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63825\")
Roger Clark has done some testing with the Canon EOS1 D Mark II in comparison to Velvia and Kodak Gold 200 print film. Like most serious photographers, he is of course using raw. His conclusions are that the Canon has 10.6 stops of dynamic range, Velvia 5 stops, and Kodak Gold 7 stops. The digital is considerably better than film. The floor is DR is determined by noise, and there can be some argument as to when image detail is overwhelmed by noise in the shadows, so there is some ambuguity to the figures.
With digital, DR is defined as full well electron capacity / read noise by CCD manufacturers, but many photographers determine the noise floor visually. In any event, with digital there is much less DR at high ISO due to the small number of electrons collected by the sensor. P&S cameras have small sensors and small full well capacities, so their DR is limited.
[a href=\"http://www.clarkvision.com/imagedetail/dynamicrange2/index.html]http://www.clarkvision.com/imagedetail/dyn...nge2/index.html[/url]
-
Metering from the midtones (or auto metering modes) can be problematic for some images, on *both* film and digital cameras. I think that those who have mastered proper metering (actaully all aspects of photo techniques) on film have some advantage when crossing over to digital. It is quite amusing to find topics like composition, etc. being extensively covered in *digital* photography books, as if these are *digital* specific.
[a href=\"index.php?act=findpost&pid=63830\"][{POST_SNAPBACK}][/a]
I agree entirely. I don't think Ansel Adams would have any trouble exposing for digital. He would merely place Zone X at the highest DN of the raw file. In his later books, he recommended against using an averaging meter, but used a 1 degree spot meter to determine dynamic range and place the tones properly. Similarly, chrome users are accustomed to exposing for the highlights.
-
Are you suggesting that DSLRs can capture a DR range equal to negative film and greater than transparencies?
[a href=\"index.php?act=findpost&pid=63825\"][{POST_SNAPBACK}][/a]
No, I am only comparing to what prints can display: traditional glossy prints can give a reflectance range of about 50:1 to 100:1, close enough to the 64:1 of 6 stops; less glossy papers give less range than that. This goes with the fact that about a 5 or 6 stops range from highlights to significant shadow detail is considered a "normal range subject", while anything more than about 6 stops requires contrast reduction at some stage to fit the "print gamut", like low contrast printing papers.
As other responses have suggested, DSLR's go comfortably beyond this 6 stop range at least at low to moderate ISO speeds: about 8 to 10 stops depending on which source you trust. Even good compact digital cameras (like the old Olympus E-10 with 5MP, 2/3" format sensor of 3.4 micron pixel pitch) have been measured as giving about 8 stops at optimum, low ISO. Slide film at least seems surpassed for DR by many digital cameras.
-
If you are metering from the midtones, you really are not exposing properly for digital ...
Also, the automated system might not be able to exclude specular highlights.
The differences in read noise at various ISOs are relatively insiginificant in most situations ...
A half stop or so of blown highlights can usually be recovered in ACR.
[a href=\"index.php?act=findpost&pid=63803\"][{POST_SNAPBACK}][/a]
I agree on everything, but maybe I have not made myself clear.
1) My whole point is that some subjects call for a different exposure settings than the CWA meter suggests: my jargon of -2, -1, 0, 1, 2 was just shorthand for "differences in exposure level from what the CW meter would say".
2) Of course, like any automated exposure system, you would at times need to change to manual mode or over-ride with a compensation setting.
3) The P. S. on my post agrees with this "in most situations" comment. I should experiment to see what borderline cases if any benefit from extra pre-amplification (higher ISO speed setting) to hold deep shadow detail that is going to be lightened in a "low contrast" print.
4) ACR saves highlights that are lost in JPEG conversion after the sensor, pre-amp and A/D convertor have handled them successfully. RAW conversion cannot save one from clipping in the pre-amp or A/D convertor (or by overflow in the sensor's electron wells themselves.)
-
i've very much enjoyed this. it's opening up areas i've not considered. i've attached three photos, because of course i have to see for myself. this is with a canon 10D, all at f/4.0, times given. 50mm f/1.4 lens. EDIT -- all shots in manual, the first two were centered on what my in-camera meter considered correct exposure.
-
4) ACR saves highlights that are lost in JPEG conversion after the sensor, pre-amp and A/D convertor have handled them successfully. RAW conversion cannot save one from clipping in the pre-amp or A/D convertor (or by overflow in the sensor's electron wells themselves.)
[a href=\"index.php?act=findpost&pid=63872\"][{POST_SNAPBACK}][/a]
I'm not sure what you mean by JPEG conversion in ACR. ACR does not convert anything to JPEG but merely converts the raw file into the designated color space in Photoshop. You can then save as JPEG if you wish.
If data are clipped in the raw file, I do not think it matters where in the above sequence the clippping took place. If the channels are clipped, they are clipped and data are lost. Some cameras do allow some headroom in the raw file. ACR highlight recovery works only if one or two channels have unclipped data--it than reconstruct the clipped data using a best guess approach. This works because the channels do not usually clip simultaneously. For example, here is a picture of a color checker taken under daylight with a Nikon digital camera, converted to linear RGB with DCRaw and with no white balance applied. As you can see the channels are not balanced and the green would blow first with overexposure, and the blue and red might remain intact.
-
i've attached three photos, because of course i have to see for myself. this is with a canon 10D....[a href=\"index.php?act=findpost&pid=63885\"][{POST_SNAPBACK}][/a]
I believe the 10D was Canon's first camera that showed a marked reduction in noise above ISO 100 (with the same exposure). Subsequent models from the 1D2 onwards to the current 30D have taken that 'apparent' noise reduction at high ISOs (whatever the true cause) even further.
-
..here is a picture of a color checker taken under daylight with a Nikon digital camera, converted to linear RGB with DCRaw and with no white balance applied. As you can see the channels are not balanced and the green would blow first with overexposure, and the blue and red might remain intact.
However, in a real scene, the channel that will clip first is likely to be the one where there is the highest degree of saturation. Landscape photographers will often find that an overexposed blue sky turns cyan (after EC in ACR) indicating that more of the blue has clipped than the green (pure RGB cyan being an equal mixture of blue and green).
I'm not sure what you mean by JPEG conversion in ACR
Bjanes, I read this comment by BJL as referring to in-camera jpeg conversion. One thing I recall about early RAW converters such as the first editions of BreezeBrowser is that they didn't do a particularly good job at recovering highlight detail. About 1/2 a stop was the most you could get, even with an overcast, predominantly grey sky. I can now get up to a whole stop in a blue sky, without serious clipping of the blue and close to 2 stops with predominantly grey areas such as a grey sky, water or grey concrete.
As you say, this might be largely due to guesswork (or better guesswork) in modern RAW converters such as ACR and C1.
-
No, I am only comparing to what prints can display: traditional glossy prints can give a reflectance range of about 50:1 to 100:1, close enough to the 64:1 of 6 stops; less glossy papers give less range than that. This goes with the fact that about a 5 or 6 stops range from highlights to significant shadow detail is considered a "normal range subject", while anything more than about 6 stops requires contrast reduction at some stage to fit the "print gamut", like low contrast printing papers.
BJL,
Whilst that is broadly true, the dynamic range issue for me is mostly about the quality of the shadows, not the fact that one might be able to faintly discern some degree of very degraded detail in the lower 3 stops of a claimed 10 stop or 8 stop DR. In the brightest parts of the image, quality is excellent till it quite suddenly disintegrates. However, in the darker parts of the image, quality just gets progressively worse, slowly descending into a total mess which is best rendered as total black on the print.
There's an implication in your above statement (although perhaps I've misread it) that we shouldn't be too concerned about the DR limitation of modern DSLRs because the reflectance DR of glossy prints is only 5 or 6 stops and therefore any digital image with a greater than 6 stop DR has to be compressed, using the wet darkroom analogy of a low contrast paper.
The eye, as it flits around, surveying a scene, is like a camera that could take multiple shots at different apertures. Unfortunately, current cameras can use only one aperture at a time to capture a scene. Multiple shots for blending purposes require a tripod and a static subject. There are many occasions when I feel the need for multiple exposures to improve DR but simply don't have a tripod. A program that could perfectly align the slight misregistration of hand-held bracketed shots would be very useful. Maybe one one already exists?
-
There's an implication in your above statement (although perhaps I've misread it) that we shouldn't be too concerned about the DR limitation of modern DSLRs because the reflectance DR of glossy prints is only 5 or 6 stops and therefore any digital image with a greater than 6 stop DR has to be compressed, using the wet darkroom analogy of a low contrast paper.
[a href=\"index.php?act=findpost&pid=63901\"][{POST_SNAPBACK}][/a]
Not quite; of course I agree that some scenes need more than an 8 stop DR.
All I am saying is that the 8-stops and more of every current DSLR at low to moderate ISO speeds is comfortably enough for scenes of low to normal subject brightness range, which are a large proportion of the total. That is, enough for scenes that are suited by "straight prints", made with no low contrast processing like pushing up of the tone curve in the shadows. Have you experienced visible noise problems in straight prints from the lower ISO settings of a DSLR? I haven't, even with my E-1, a camera much accused of having "too much noise".
-
However, in a real scene, the channel that will clip first is likely to be the one where there is the highest degree of saturation. Landscape photographers will often find that an overexposed blue sky turns cyan (after EC in ACR) indicating that more of the blue has clipped than the green (pure RGB cyan being an equal mixture of blue and green).
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63893\")
It takes quite a bit of overexposure to blow out blue sky, since neither the luminance nor saturation of clear blue sky is that high. Much more likely, is blowing out of the clouds in a blue sky as shown in the shot of the Golden Gate Bridge in Bruce Fraser's article. With the histogram in ACR it is easy to see which channels are blown. In my experience with ACR and my shooting, the blown channels most often show as white in the histogram, as in Bruce's example, indicating all three channels are blown in the white balanced image. Your mileage may vary, but the essential point is that highlight recovery requires data in at least one channel.
[a href=\"http://www.macworld.com/2005/03/secrets/marcreate/index.php]http://www.macworld.com/2005/03/secrets/marcreate/index.php[/url]
Bjanes, I read this comment by BJL as referring to in-camera jpeg conversion. One thing I recall about early RAW converters such as the first editions of BreezeBrowser is that they didn't do a particularly good job at recovering highlight detail. About 1/2 a stop was the most you could get, even with an overcast, predominantly grey sky. I can now get up to a whole stop in a blue sky, without serious clipping of the blue and close to 2 stops with predominantly grey areas such as a grey sky, water or grey concrete.
[a href=\"index.php?act=findpost&pid=63893\"][{POST_SNAPBACK}][/a]
Again, bolstering my point about the sky blowing highlights and reinforcing my point about white balance multipliers in the color channels. However, I still do not see how JPEG is related to highlight recovery in RAW. Moreover, highlight recovery does not work with JPEG.
-
Since so many are chiming in with their opinions on this topic, it is perhaps practical to add mine. I think that Rags has the right idea but I also think the placement of highest important value should be relative to the number of tones the output target can support. A zone V placement may or may not be appropriate for some papers and the scene's range should be kept in mind.
The goal in digital photography as I see it, is to distill from the total number of tones our capture device records, the correct number of tones for the target we've chosen. This correct number of tones needs to have a progression that illustrates the emotion the shooter wants to imbue his image with. This means the total contrast has to be correct and the local contrast has to be correct. Perhaps the best conductor of tones we've ever seen is Ansel Adams. Would Ansel make a grade 1 neg for grade 5 paper? Not likely, at least as far as physics is concerned. A grade 1 neg has way more tones than grade 5 paper can render. Consider that the expose right crowd is making a grade 1 "neg"/file for every scene and every target. The mantra is "get all the data". When faced with the idea that you will never be able to render all these tones on some reflective targets, what's the point? Why not capture the number of tones that you want?
This capturing of tones is problematic, when faced with a histogram for a report of where the highest important tone to you, resides within that histogram. Good luck finding it in the data stacks. The reality of the histogram suggests you cannot place an important value; all you can do is make sure you don't saturate the image, so that's the ritual. Give the scene as much exposure as you can, and head for Photoshop.
Rags' contention that less exposure may be better has it's roots in manufacturing a specific number of tones for a picture rendering. While zone V is drastic for some scenes and targets, it's not all that unbelievable for some capture devices. Why? Let's compare the number of tones in 12 bits (4096) with the signal-to-noise ratio of a high end camera back (4000:1). We've got approximately 4000 units in each comparative and the equation suggests that the noise in 12 bits is approximately 1 bit. In 12 stops of range, zone I has noise (0 is black) and the others are noise free.
For a normal scene (for example, let's say it's a 6 stop scene), one could expose right, place the highest important value way high with the histogram and no zone in the scene has noise. One could also expose the highest important on zone VII, and no zone in the scene has noise.
Editing these two pictures will be different. If the highlight needs contrast, you're stuck with Photoshop's special features like Hightlight & Shadow; since you can't increase highlight contrast without depressing the curve below the highest important value-- there's no room to map the high value upward with more contrast. Depressing the curve depresses tones while it adds contrast below the highest important value, and it flattens the shadow. If you're thinking that here's a great way to sell a software feature like Highlight & Shadow, you're right. Of course there's always the layer masking, selection based, high bit, gotta have a $2,000 monitor, post production gymnastics that the expose-to-the-right crowd is proposing, but does that fit the regular Joe's budget? If it does, you may be a proponent of expose-to-the-right.
If we look at the zone VII exposed highest important value, there's flexibility in mapping this placed value upward to the print target's highlight. You can add contrast, or you can smoothen the adjacent values; it's your choice. And, all the tones below are stretched in proper context, appropriately emotional, with the resultant tonality quite filmlike and unlike the overprocessed, artifical, and invasive edits common today. With many shooters, more edits, layers, masks, and complexity equates to a better picture. I don't agree, I'd still rather see one of Ansel's best, compared to the modern digital ethic's "best". A trip through several of today's modern photo rags will illustrate my point on overprocessed imagry. Very little of it looks real.
If you don't possess a 4000:1 capture device, you may be relegated to "expose to the right". If you have one of these better performing devices, do some testing and see if you like the result. The pictures will look different, only you can decide if you like the result.
We're early on in the digital photography evolution, we've been at it for only 20 years. At first we didn't have enough tones. Now, we have more than we can print. Between these ethics, we've struggled for more tonalty and now that we have it, we've become complacent by ritualizing the sending of too many tones. As we progress, it's practical that we'll all discover that the true craft of all this technology in making truly emotive images, is best acheived by making the right number of tones for what you want to visually say.
Richard Chang
-
Would Ansel make a grade 1 neg for grade 5 paper? Not likely, at least as far as physics is concerned. A grade 1 neg has way more tones than grade 5 paper can render. Consider that the expose right crowd is making a grade 1 "neg"/file for every scene and every target. The mantra is "get all the data". When faced with the idea that you will never be able to render all these tones on some reflective targets, what's the point? Why not capture the number of tones that you want?
For a normal scene (for example, let's say it's a 6 stop scene), one could expose right, place the highest important value way high with the histogram and no zone in the scene has noise. One could also expose the highest important on zone VII, and no zone in the scene has noise.
Editing these two pictures will be different. If the highlight needs contrast, you're stuck with Photoshop's special features like Hightlight & Shadow; since you can't increase highlight contrast without depressing the curve below the highest important value-- there's no room to map the high value upward with more contrast. ? If it does, you may be a proponent of expose-to-the-right.
Richard Chang
[a href=\"index.php?act=findpost&pid=63972\"][{POST_SNAPBACK}][/a]
Richard,
A thoughtful post, but I don't think your analogy to Adam's zone methods is quite correct. If the dynamic range of the subject is equal to that of the film or sensor, one has no exposure lattitude. With digital you should place the highlights just short of clipping--any less exposure and the shadows will be clipped. With negative film, you should place the shadows requiring tonal texture on Zone II. The highlights with texture will then fall on Zone IX.
With a short scale subject where the dynamic range of the scene is less than that of the sensor or film, you have some lattitude in placement. With film, you can place the shadows higher than with the full scale subject, and the highlights will still be on the linear portion of the characteristic curve. You might place the shadows at Zone III for additional detail, but further additional exposure increases grain, decreases acutance and is not recommended by Adams. Depending on your visualization, you might also consider expansion for the short scale subject.
With digital and a short scale subject, it is best to place the highlights just short of clipping so as to maximize the number of tones and signal to noise ratio. If you expose Zone VII at Zone X, you then use -3 EV compensation in ACR. The contrast does not change.
Of course digital response is linear whereas film is log and digital lacks the knee and shoulder of film and clips abruptly. When gamma correction is applied to digital, the resulting curve then resembles the characteristic (H&D) curve of film.
-
The mantra is "get all the data". When faced with the idea that you will never be able to render all these tones on some reflective targets, what's the point? Why not capture the number of tones that you want?
[a href=\"index.php?act=findpost&pid=63972\"][{POST_SNAPBACK}][/a]
Simply because then you are locked into the scene interpretation you made when capturing and reduce the flexibility to reinterpret with post processing.
And while you may think post processing is less desirable, I would argue that an artist will do whatever it takes to get the image to look the way they want.
You choose to talk about Ansel Adams. Do you know how he worked? Do you realize that he would develop by inspection back when the films he used were orthochromatic? He would use N+ and N- processing to control the exact density range of his negs. Do you honestly think he printed in the darkroom without doing things like dodge/burn, hot processing, local finger agitation?
No, those were the tools of the day. Those techniques were his post-processing, his Photoshop. The "Zone System" has progressed far afield from the limitations Adams faced in his darkroom.
There is nothing particularly nobel about engaging in photography without post-processing. It's an arrogance some engage in, while others see the end results as what is important. True, depending on your intention, it can be less than honest if you engage in deceptive alterations outside of the context of the intent-such as journalism or documentation. I'm not limited by those restrictions.
A print hanging on a gallery wall doesn't tell the viewer how it's been modified by some post-processing, not if you know what you're doing. A print just shows what the artist deems important. The viewer really has no further knowledge about how the image was made. As it should be.
You all are free to do whatever the heck strikes your fancy, but if you follow Rag's advice and expose for Zone V, you are wasting opportunity, reducing flexibility, locking in the scene interpretation and do not really have a grasp of the process and the craft that is digital capture.
-
I don't see much point agonizing over the zone system when doing digital photography. We'll get the best results we can using the technology the way it should be used, and the way it should be used depends on its technical characteristics. If we can end-up with less noisy appearance and smoother gradations by exposing to the right and then deriving the luminosity we want using curves and layer-masking so be it - that's the way to go.
-
I've shot in excess of 100,000 frames with DSLRs (Canon 1Ds, 1D-MkII, and 10D), and about 10,000 frames with P&S digitals, and I have yet to find a subject that materially benefits from placing important highlights (those that need detail for the image to "work") more than a half stop below clipping. As long as no important highlights are clipped, increasing exposure always improves noise levels and shadow detail, resulting in a RAW file that offers the greatest flexibility in processing. If a dark "look" to the image is desired, exposing to the right and converting the RAW with a negative exposure setting can deliver an image with identical tonality to one given less exposure, but the image with greater exposure will be much cleaner and will require significantly LESS post-processing than the frame with less exposure.
-
Have you experienced visible noise problems in straight prints from the lower ISO settings of a DSLR? I haven't, even with my E-1, a camera much accused of having "too much noise".
[a href=\"index.php?act=findpost&pid=63956\"][{POST_SNAPBACK}][/a]
BJL,
Yes I have. It's probably a consequence of living in a sunny clime where brightness levels are greater than in northern Europe. Current DSLRs are inadequate for many rainforest shots as well as sunset shots in general.
We should bear in mind that 'artists' (a phrase I'm not inclined to use, "Woohee! I am an artiste ") will strive to make use of the limitations of the medium to the best advantage. This might include using generally obnoxious grain in film to create an effect of austerity, or whatever. It might include using the camera's limited dynamic range to reduce distracting detail and induce an aura of mystery through the creation of black shadows. Michael R is very good at this. His recent shots from Nambia emphasise form and contrast to great effect.
The resulting tradition of how great photographers have handled these limitations results in a style to be imitated by less experienced newbies; the portrait with one side of the face in deep shadow, for example.
We get used to seeing such examples as the pinnacle of photographic achievement, perhaps not realising that they were essentially an 'artistically' successful reaction to the limitations of the medium.
But supposing we want the camera to produce what the eye sees? For some people, that is a given. But the more sophisticated of us realise that it's not a given. There are many instances when the camera simply fails to deliver what the eye sees. Dynamic range limitations of the camera are one of them.
I'm going to illustrate this point with a few shots of a sunset which are not particularly great shots but were taken shortly after I discovered that my newly purchased 24-104 IS Canon lens had a flare problem. I set about searching for this problem. I didn't find it, but the following shots are part of the result of my efforts to find it.
The first shot below is a straight conversion with shadows at zero, EC at minus 1 stop, contrast at -50 and luminous and color NR at maximum. I left brightness at default 50 so you could see the thing. Here it is.
[attachment=519:attachment]
I'm sure you will all appreciated that this image is definitely not what the eye sees. It needs working on. The tree on the left is a distraction. Let's crop it out. The foreground needs lightening (for what the eye sees). I know it was never that dark when I took the shot.
Here's the reult.
[attachment=520:attachment]
Now here we're into endless artisitic discussions about appropriate levels of brightness. I'm going to cut through that and deliver a crop of part of the image which 'approximates' what I saw when I took the shot. As you can see, it's a very degraded image.
[attachment=521:attachment]
ps. I should add a couple of points which I forgot to mention in the original post. Minus 1 EC was in relation to a Prophoto color sapce. ARGB would have necessitated a -2 EC correction in ACR to reduce the spike at the right side of the histogram.
The final shot is only a poor approximation of what I saw.
-
With regard to the comments by Jeff Schewe:
Why not capture the number of tones you want?
> Simply because then you are locked into the scene interpretation you made when capturing and reduce the flexibility to reinterpret with post processing.
Why is it you think I need to reinterpret? Why can't I decide exactly how I want my image to render, on the target I choose? Why can't I decide how to render my vision? Why to I have to make the compromise exposure for what I want, simply to allow a reinterpretation that I'm not interested in?
>And while you may think post processing is less desirable, I would argue that an artist will do whatever it takes to get the image to look the way they want.
I never said that post processing is less desireable, I suggested that I don't appreciate the over processed look of some of the expose-to-the-right imagery. I do post processing, layer adusts, masks and the like. I even did the Schewe edge sharpen routine until the new Photoshop sharpening showed up.
>You choose to talk about Ansel Adams. Do you know how he worked? Do you realize that he would develop by inspection back when the films he used were orthochromatic?
I do know how he worked and I did development by inspection in the early 70's too. I'll defy you to accurately see the proper total development of a finely detailed reflectance on a neg, in the low level safe lighting that Ansel inspected in. It's not like checking the densitometer in Photoshop. How Ansel's early development transpired is taking my point on a tangent and it's not relevant to my opinion. Perhaps you can comment on some of Ansel's Polaroid P/N 55 imagery from Yosemite in the winter, what kind of tangent can you present here? Ansel gave up on inspection development as soon materials got consistent. Are you trying to suggest that the Tri-x and HC110 development table of times many of us used were articulated by Ansel with inspection development? Ansel's images from the later portfolios didn't involve inspection nor did they need it. To characterize Ansel's work as inspection development work is a stretch to me.
>Do you honestly think he printed in the darkroom without doing things like dodge/burn, hot processing, local finger agitation?
Of course not: I own The Print (in fact, I own the whole series). I know about flagging contact printers, printing with KodakAzo and Agfa Contactone, jiggle machines, Codelights, metronomes, DeBeers formulas, and all the other stuff he wrote about.
>There is nothing particularly nobel about engaging in photography without post-processing. It's an arrogance some engage in, while others see the end results as what is important.
I agree that there is nothing noble about eschewing post processing. I also agree that it's an arrogance that some engage in. And I do belive that what the audience sees is what is important, which is why I don't expose-to-the-right. I choose to handle the scenes I shoot, differently. And I choose to not have to take the image editing path that is recommended (and is sometimes required) by the expose-to-the-right crowd.
>A print hanging on a gallery wall doesn't tell the viewer how it's been modified by some post-processing, not if you know what you're doing. A print just shows what the artist deems important. The viewer really has no further knowledge about how the image was made. As it should be.
Amen to that; we agree. I know you'll do what you espouse and I know I'll do what what strikes my fancy. I don't want my pictures to look like yours, any more than you want your pictures to look like mine. We both think we know what we're doing, and that's fine by both of us.
>You all are free to do whatever the heck strikes your fancy, but if you follow Rag's advice and expose for Zone V, you are wasting opportunity, reducing flexibility, locking in the scene interpretation...
I will do what strikes my fancy. Rag's advice to expose for Zone V won't work for every scene and I don't agree that one should. But less than expose-to-the-right works best for me on a number of levels. My choosing to lock scene interpretation is deeply rooted in Ansel's teaching; it worked for me when I shot film and I embrace it with digital. I don't want flexibility, I want the best rendering for my previsualization and I want to lock in my scene interpretation for my chosen target.
>...and do not really have a grasp of the process and the craft that is digital capture.
As a technical representative of a camera back maker (MegaVision, Inc.) going back 10 years, I'll believe that I know a fair amount about the process and the craft that is digital capture. Besides working for MegaVision, I've run imagesetters and made film based proofs on DuPont, Kodak, and Fuji proofing systems with the film. I used color management with Efi's Cachet prior to the notion of device independent color. I've been around this digital capture business since 1992 and I've used Ansel's concepts to teach digital on 6 continents.
When I first heard you at a trade show, you were telling everyone that digital capture wasn't good enough and that film and scanning was the only way to go. Digital tonality and the way we handle tones hasn't really changed since then, but we do have new Photoshop tools and we now have your notions of what process and craft are all about.
In the old days of film, some guys could shoot Agfa, some could shoot Minoltas, and some could shoot Cambos and Speedos, even though they weren't mainstream. All that counted was the picture. We ought to hold our image making to the same standards; what the picture looks like should be what's important.
Richard Chang
-
Why is it you think I need to reinterpret? Why can't I decide exactly how I want my image to render, on the target I choose? Why can't I decide how to render my vision? Why to I have to make the compromise exposure for what I want, simply to allow a reinterpretation that I'm not interested in?
Well, I for one would not want to be locked into a target in the field. How are you evaluating the resulting capture in the field? I certainly hope your determination isn't based on the camera's LCD, ala "chimping" and a histogram only tells you the image stats, not what the heck it looks like...me? I wait until I get back to the studio and use a well profiled display to make the final determination of what the image SHOULD look like...and I'll be the first to admit that upon first seeing an image on my display, I may just want to change it-heck, I'll almost ALWAYS want to change it!
As long as I've achieved the maximum usable data capture I have that flexibility to interpet the scene any way I want in post. That is indeed what the whole digital revolution is all about for me, absolute total control over the final printed image. So, why would I want to limit what I can do with an image right at the very start of the process-at the point of capture. It's the primary reason I shoot raw instead of jpg. Your way of limiting data capture at the point of exposure is akin to shooting jpg. You bake a tone curve in at capture? I don't think so...at least not for me.
To characterize Ansel's work as inspection development work is a stretch to me.
I didn't, but you may have taken it that way-which is wrong. I specifically said he developed by inspection when shooting orthochromatic films...he gave up instpection development when he switched to panchromatic films. Safelights for panchromatic films are not really useful for anything other than seeing a very dim spot in space. :~)
When I first heard you at a trade show, you were telling everyone that digital capture wasn't good enough and that film and scanning was the only way to go. Digital tonality and the way we handle tones hasn't really changed since then, but we do have new Photoshop tools and we now have your notions of what process and craft are all about.
Pretty sure that was in the early 90's, right? And digital capture back then -WAS- so limiting that film was still far superior to digital capture. A lot has changed in 10 years don't you think Richard?
The first digital capture I did was with a Megavision back (or was it a Carnival back?) that I rented in 1996. I was able to turn the job around-the final halftone reproduction of 4x5 @ 150LPI in an incredably short time (the reason we went with digital). I think it was a 2K capture which with cropping was just good enough.
When Canon came out with the 1Ds I did an article for PEI magazine which proved to me that with the exact same lens, comparing 35mm film and 11MP capture, the digital capture out resolved the film, had better dynamic range and better color. That was the last time I really ever shot film with the exception of certain large format stuff that needed extensive camera movements.
I wasn't sure it was you Richard, when you first posted...glad to see you're still around! Now please tell me you haven't drank Rag's Cool-Aid, ok?
:~)
-
Well, I for one would not want to be locked into a target in the field. How are you evaluating the resulting capture in the field? I certainly hope your determination isn't based on the camera's LCD, ala "chimping" and a histogram only tells you the image stats, not what the heck it looks like...
Actually, I'm using an OQO computer running MegaVision's Photoshoot 4 capture software. I load a paper target into our Color Coded Light Meter (fancy term for gamut alarm meter) which illustrates in colors, those scene pixels that constitute the renderable endpoints of my paper-- red is blown and yellow is a just-separable high. Dark blue is plugged on the chosen paper and light blue represents the lowest separable values. Other colors denote 3/4, midtone, and quarter-tone. There's a histogram but it's clearly inferior to Color Coded Light Metering. The OQO also runs CS2, so I have that as well if I want to check anything on the spot. It all fits on the back of my camera-- http://www.mega-vision.com/products/1shot/1shot.htm (http://www.mega-vision.com/products/1shot/1shot.htm). The display is 3.5 x 5 inches and it can easily be seen in direct sunlight.
>As long as I've achieved the maximum usable data capture I have that flexibility to interpet the scene any way I want in post. That is indeed what the whole digital revolution is all about for me, absolute total control over the final printed image.
For many workflows, that's a fine approach. Ansel taught me to make a decision about what I saw (previsualization) and take whatever steps were necessary to acheive that previsualization. This style isn't what you're doing, but it isn't broken either-- it's just a different style.
>So, why would I want to limit what I can do with an image right at the very start of the process-at the point of capture.
Because you've decided exactly what you want to do? According to your text, that doesn't happen and you decide later. I don't suffer that lag in decision making.
What you do decide to do, is directed by the look of the monitor, the eventual success of your edits, and your remberance of the scene, which may be recent or it may not be recent. Why wouldn't I want to encode my vision behind the camera, into my picture? Why do I have to wait till I get to a calibrated display to make decisions? Your workflow works for you but it isn't spontaneous, at least not in a traditional photographic sense. It sounds like more of a reaction to what you end up with back at the studio.
Of course we all bracket our exposures, it would be infantile not to. Which brackets do I end up using most of the time?-- the one that has caused me to express my opinion to this thread.
>It's the primary reason I shoot raw instead of jpg. Your way of limiting data capture at the point of exposure is akin to shooting jpg. You bake a tone curve in at capture? I don't think so...at least not for me.
I don't shoot jpg. MegaVision instituted RAW capture in 1996 with our then named MEGA file. We don't bake in a tone curve either, we're always shooting raw but you get to see the development intention in Photoshoot, expressed in tenths of a stop for push and pull, and tenths of a logarithmic density value in our DR. Change the intention and see the result on the paper with the Color Coded Light Meter. Shooting the picture saves the raw (our raws are .dng's) to disk; you can use our developing or go to ACR.
Luminance and chromnance are handled separately in Photoshoot, we don't remap ratios of RGB. And we're fully savvy with regard to ICC profiles, working spaces, and rendering intents. We also have IPTC, Metadata, and searchable keyword databasing in Photoshoot.
Regarding Ansel's inspection development:
> I specifically said he developed by inspection when shooting orthochromatic films...he gave up instpection development when he switched to panchromatic films.
If you check my previous commentary on this, you'll see I agree. I'll believe that he made way more pictures with panchomatic emulsions than he did with ortho's. My comments tried to address that Ansel didn't do a lot of development by inspection and that he dropped it like a hot rock when panchromatic emulsions came about. I'll also offer that with the panchromatic emulsions, the zone system became more difficult because the film had much better contrast (it became a relatively modern emulsion as we know it), and Ansel used it to shoot images like Mt. Williamson, the Grand Tetons, and Mt. McKinley. Ansel's best tonal renderings weren't made with ortho films. My comments about Ansel were directed to this imagery-- I didn't mention development by inspection because I don't think it has merit today.
>Safelights for panchromatic films are not really useful for anything other than seeing a very dim spot in space. :~)
I think I mentioned something to that effect.
Meeting up the first time:
>Pretty sure that was in the early 90's, right? And digital capture back then -WAS- so limiting that film was still far superior to digital capture. A lot has changed in 10 years don't you think Richard?
Yes, it was early on when you talked on a panel regarding copyright-- possibly at the Javits.
Far superior is relative, and although a lot has changed, some things haven't. In the early days of digital, we were restricted to smaller image areas and smaller target renderings. When we shot for catalog illustrations (less than full page) we had just adequate resolution with a 4 megapixel 3 pass. Photoshop was amazing then too. Yes we had to light carefully, but we could manage tones as well then as we can now-- it was a narrower envelope then but the picture rendering was as good as it is now, at least in process color on press. I do like today's extra resolution and improved signal-to-noise, and the ability to have the computer on the back of my camera gives me capability I didn't have then.
>I wasn't sure it was you Richard, when you first posted...glad to see you're still around! Now please tell me you haven't drank Rag's Cool-Aid, ok?
I am still around. And no, I haven't drunk Rag's Cool Aid, but I will say that I've given less exposure to my work; based on the target, the scene contrast, and how I want to render the scene. Yes I have a cutting edge s/n ratio in my E5 back which allows me this luxury, and I do expose-to-the-right with my Canon Powershot. I take the path my equipment allows me to take.
I can see that you're still as feisty as always, keep it up! MegaVision is doing Photokina and Photoplus at the Javits this year; if you're around, stop in and see us.
Richard Chang
-
I agree entirely. I don't think Ansel Adams would have any trouble exposing for digital. He would merely place Zone X at the highest DN of the raw file. In his later books, he recommended against using an averaging meter, but used a 1 degree spot meter to determine dynamic range and place the tones properly. Similarly, chrome users are accustomed to exposing for the highlights.
[a href=\"index.php?act=findpost&pid=63841\"][{POST_SNAPBACK}][/a]
What we need is a book or tutorial titled "Digital Cameras for Experienced Film Photographers", which focuses on how digital DR, metering, noise, ISO, etc. differ from film, and how certain photo techniques remain unchanged from shooting film to digital.
-
No, I am only comparing to what prints can display: traditional glossy prints can give a reflectance range of about 50:1 to 100:1, close enough to the 64:1 of 6 stops; less glossy papers give less range than that. This goes with the fact that about a 5 or 6 stops range from highlights to significant shadow detail is considered a "normal range subject", while anything more than about 6 stops requires contrast reduction at some stage to fit the "print gamut", like low contrast printing papers.
As other responses have suggested, DSLR's go comfortably beyond this 6 stop range at least at low to moderate ISO speeds: about 8 to 10 stops depending on which source you trust. Even good compact digital cameras (like the old Olympus E-10 with 5MP, 2/3" format sensor of 3.4 micron pixel pitch) have been measured as giving about 8 stops at optimum, low ISO. Slide film at least seems surpassed for DR by many digital cameras.
[a href=\"index.php?act=findpost&pid=63869\"][{POST_SNAPBACK}][/a]
Thank you and others for commenting on digital DR. All your posts seem to say that digital DR is a function of ISO and noise. That's a relationship not apparent on film.
-
It takes quite a bit of overexposure to blow out blue sky, since neither the luminance nor saturation of clear blue sky is that high. Much more likely, is blowing out of the clouds in a blue sky as shown in the shot of the Golden Gate Bridge in Bruce Fraser's article. With the histogram in ACR it is easy to see which channels are blown. In my experience with ACR and my shooting, the blown channels most often show as white in the histogram, as in Bruce's example, indicating all three channels are blown in the white balanced image. Your mileage may vary, but the essential point is that highlight recovery requires data in at least one channel.
http://www.macworld.com/2005/03/secrets/marcreate/index.php (http://www.macworld.com/2005/03/secrets/marcreate/index.php)
[a href=\"index.php?act=findpost&pid=63970\"][{POST_SNAPBACK}][/a]
Recovering highlights wih only one or two unclipped channels is not that easy. Often, such highlights are very saturated and out of (printer profile) gamut. At least that's my experience with film scans.
While PS RAW histograms display separate channels, do in-camera histograms do so also? If not, won't ETTR relying on histograms result in potential clipped channels?
-
Recovering highlights wih only one or two unclipped channels is not that easy. Often, such highlights are very saturated and out of (printer profile) gamut. At least that's my experience with film scans.
While PS RAW histograms display separate channels, do in-camera histograms do so also? If not, won't ETTR relying on histograms result in potential clipped channels?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=64076\")
I haven't scanned any film since I went digital, so I have no experience recovering highlights in that situation and did not even know it was possible. Fortunately, you still have your original and can re do it.
The single histogram on most cameras is unfortunately a luminance histogram, not what Adobe calls an RGB histogram. In the luminance histogram, the RGB colors are weighted according to the sensitivity of the eye, and the red and blue channels do not get much weighting.
[a href=\"http://www.cambridgeincolour.com/tutorials/histograms2.htm]http://www.cambridgeincolour.com/tutorials/histograms2.htm[/url]
With a lumonsity histogram it is very easy to blow a blue or red channel. Here is an example from my own experience. I exposed this red flower to the right according to the lumonsity histogram on my D70. In Photoshop, the lumonsity histogram looks fine (look at the black and white histogram) and resembles what I saw on the camera. However, the RGB histogram shows clipping of the red channel. My new D200 has separate RGB histograms which would avoid this problem, as do the professional Nikons and Canons. If a camera had only one histogram, one would think that it should be what Adobe calls the RGB histogram so as to avoid this problem.
[attachment=522:attachment]
[attachment=523:attachment]
BTW, the camera histogram does not show the RAW channels, but those of the JPEG preview, which has been converted to the assigned color space with the current camera settings. If you use ProPhotoRGB in ACR, the red clipping does not occur in the wider color space. Unfortunately, this space is not available on most cameras. (These screnshots are posted as GIF to better show text, hence the posterization in the colors)
[attachment=524:attachment]
[attachment=525:attachment]
-
It takes quite a bit of overexposure to blow out blue sky, since neither the luminance nor saturation of clear blue sky is that high. Much more likely, is blowing out of the clouds in a blue sky as shown in the shot of the Golden Gate Bridge in Bruce Fraser's article. With the histogram in ACR it is easy to see which channels are blown. In my experience with ACR and my shooting, the blown channels most often show as white in the histogram, as in Bruce's example, indicating all three channels are blown in the white balanced image. Your mileage may vary, but the essential point is that highlight recovery requires data in at least one channel.
[a href=\"index.php?act=findpost&pid=63970\"][{POST_SNAPBACK}][/a]
Golden Gate Bridge? I'm talking about clear, unpolluted deep blue sky with the odd cloud that varies from dark grey to brilliant white.
A Chinese gentleman who lived about 2500 years ago is reputed to have said, but probably never did say, that a picture is worth a thousand words. There's no blue sky in S.E Queensland at the moment of writing this, so I can't demonstrate my point with a picture, and since I'm so careful about this issue of overexposure of sky, I can't off-hand find a good example in the limited archives at my disposal at this location.
If one is doing a cloud study and exposing for the cloud highlights, then that's fine. You'd possibly get some interesting, dramatic shots showing every nuance of cloud detail, and a very dark blue sky. However, if a bit of the scene was terra earth in the foreground, you'd be stuffed as regards extracting noise-free detail.
We generally expect white clouds to be clipped to some extent, and the clipping seems (in my experience) to simply produce greater areas of pure white without any obnoxious color shift, and is quite acceptable.
A clipped blue sky seems to behave differently when trying to recover the highlights in ACR with a minus EC setting. Beyond about -1 EC, the blues in the brighter parts of the sky gradually turn cyan and in extreme cases a muddy grey. Basically, a total mess which requires a lot of work in Photoshop to correct.
Okay?
-
Golden Gate Bridge? I'm talking about clear, unpolluted deep blue sky with the odd cloud that varies from dark grey to brilliant white.
We generally expect white clouds to be clipped to some extent, and the clipping seems (in my experience) to simply produce greater areas of pure white without any obnoxious color shift, and is quite acceptable.
A clipped blue sky seems to behave differently when trying to recover the highlights in ACR with a minus EC setting. Beyond about -1 EC, the blues in the brighter parts of the sky gradually turn cyan and in extreme cases a muddy grey. Basically, a total mess which requires a lot of work in Photoshop to correct.
Okay?
[a href=\"index.php?act=findpost&pid=64092\"][{POST_SNAPBACK}][/a]
Sorry, Ray, but I don't get your point. In the example, Bruce wanted to bring out some texture in the clouds, not merely represent them as white blobs. Sounds reasonable to me.
In my experience a clear blue northern sky generally has a luminance reading on the light meter very similar to that of the overall scene, which is assumed to have a reflectance of 12-18% in most metering schemes. Ansel Adams describes a clear north sky as rendered by panchromatic film as Zone V and I don't think it is much different with digital.
Some digital cameras run the sensor slightly above its linear range, so channel clipping will not be uniform across all channels, and a linear control such as the exposure control in ACR may not correct for this and there will be a color shift. Furthermore the white balance is a factor. These considerations are explained by Bruce Fraser in his ACR book.
If it is your misfortune to have drab overcast skys in your location, overexposure can be a problem. If you expect more than -1EV of highlight recovery in your situation, I don't think you are being realistic. You could watch your camera histograms and avoid blowing channels or else use a graduated neutral density filter, HDR exposures, etc.
-
Of course we all bracket our exposures, it would be infantile not to. Which brackets do I end up using most of the time?-- the one that has caused me to express my opinion to this thread.
[a href=\"index.php?act=findpost&pid=64034\"][{POST_SNAPBACK}][/a]
Richard,
Would any of these bracketed shots be an ETTR exposure, by any chance?
Your method of shooting seems very elaborate and time consuming. 3 bracketed shots around the centre of your very precise measurements seems a good insurance measure and it would hardly add to the over all time of preparing the shot.
I agree in principle, if you are so absolutely sure of your methods and your vision, and you are not interested in changing those at a later stage, then following the tradition of Ansel Adams might serve you well.
Your method seem to me to be the equivalent of shooting transparencies, but in digital mode. You've got a final result which inspires cofidence that that was what you envisioned. The hues and tones are right without relying upon memory.
I understand that approach. But it wouldn't suit me because I rely just upon the camera (a 5D) with its in-built processes, and most of the time I don't have the luxury of hanging around to take precise measurements, even if I had the ancilliary equipment.
Cheers!
-
Richard,
Would any of these bracketed shots be an ETTR exposure, by any chance?
No way!; I have Color Coded Light Metering in my Photoshoot capture app. I've got a light meter of each and every pixel, all 22 million of them. I can sample (from point sample up to 21x21) and I can see each pixel's zonal value in the densitometer, and I can see important endpoints as colors. If I change the developing intention's luminance remap, I see the colors change. Whether I choose to develop in ACR or Photoshoot, I can optimize the exposure for the developing I'll want to use. Why not be able to change the exposure of the scene, prior to the developing you want? Sure you can change the deveoper in ACR, but why limit yourself? You can have both exposure and developing adjustments on your tripod if you have the capability.
The developing I want has it's roots in film. I know what a normal development does to lows, mids, and highs. I know what a normal-minus and a normal plus development looks like too. When I see a scene that needs a plus in the highs, I'll try to give an exposure that allows me the most flexibilty in getting the rendering I want. I've tried giving lots of exposure and using Photoshop's H&S, layer masks, etc. For me, giving less exposure and curve mapping up with more contrast, is the way I want to work. There's a big advantage for me, if the scene still fits noiselessly in my camera's recording-- when I map up, all the tones below the mapped point will move in context. Give max exposure and for some scenes, you'll be heading for selections and Photoshop gymnastics to save your shot. Bear in mind that Photoshop has some highly developed gymnastic routines which have been marketed with enthusiasm and great fervor. Always great for Adobe but but not maybe so great for you.
>Your method of shooting seems very elaborate and time consuming. 3 bracketed shots around the centre of your very precise measurements seems a good insurance measure and it would hardly add to the over all time of preparing the shot.
Time consuming? If you've shot much outdoors you'll know that the light dictates how much time you're able to spend. Ansel didn't spend much time on Moonrise, Hernandez, NM because he knew the light wasn't going to stay on the headstones for long. I wouldn't either, in a similar situation. The time I spend shooting depends on what kind of time I want to spend. No one is going to make you take a lot of time. If you do want to spend time, go for it, it's your time. Of course, as you hone any skillset, it takes less time. I've been doing this since late 1998 when we got the first Philips sensor.
I like to think that most shooter's work visually communicates the time they spent. Do you need to spend more time on your images?, or would less time be more appropriate?
>I agree in principle, if you are so absolutely sure of your methods and your vision, and you are not interested in changing those at a later stage, then following the tradition of Ansel Adams might serve you well.
It does and I won't change my workflow as long as I have the s/n that allows it. You'd need to see some prints to find out if you agree as well. Trouble is, you won't find a lot of shooters espousing the placement of highest value according to the paper's density range and highest separable value. Most folks ritually give as much exposure as possible, and I do too, when I shoot with gear that doesn't allow less exposure.
>Your method seem to me to be the equivalent of shooting transparencies, but in digital mode. You've got a final result which inspires cofidence that that was what you envisioned. The hues and tones are right without relying upon memory.
We might find value in considering that a digital image isn't a negative, it's a positive. Ansel taught us to expose for threshold and develop for the other end of the scale. Threshold for digital, being a positive, is the start of density on the paper or monitor. I teach my customers that threshold is what we expose for and that shooting the number of tones (whenever possible) that match the measured DR of the paper will allow a perfect rendering.
>I understand that approach. But it wouldn't suit me because I rely just upon the camera (a 5D) with its in-built processes, and most of the time I don't have the luxury of hanging around to take precise measurements, even if I had the ancilliary equipment.
Maybe not, but it doesn't have to take a lot of time. If I had a 5D, I'd test it by shooting a test scene; place the highest important value on zone IX, VIII, VII, VI, and V. Use the spot function in the camera's reflected meter mode to choose your exposures. As to precise measurments, I'll assume that you remember how to use an in-camera meter-- it reads only the reflectance in the metered area. Open the pictures in CS2 and look at the results. CS2's automatic function will do it's best to make the shot look the way Adobe thinks you want it. Check out the highs, and the lows. Depending you your camera's performance and the original scene contrast, the lows will be noisy, or they'll be OK. The highs will be more textured, as you give less exposure. Of course you may alter Adobe's notion with ACR's tools.
Even if you don't use the technique, it might be nice to quantify your camera's performance, handy for eval against future gear you obtain, if you standardize your test scene; I use sunlight and a 24 patch Colorchecker.
Whether you give this test a shot is up to you. If you use the technique, that's up to you too. You can objectively take a look at what the advantages and disadvantages engender, understanding that as the performance of the device improves, you gain the luxury of lower placment of highest important value. It's my opinion that rendering a correct number of tones to our target is more emotive than rendering too many. It why we like Velvia in flat light. Tonal craft is what really good shooters manage. Luckily, you can manage in more than one style, just choose the style that works for you and your equipment.
Richard Chang
-
OK, Richard...I don't get it.
You say (and I very much agree) to expose to retain textural highlights in a digital capture. Fine. So, where does this "reduce the levels of tones captured" come into play?
If you properly expose to just maintain textural highlights with no clipping, I would consider that a "good exposure". It would be an "expose to the right" senario if the scene contrast range was within the dynamic range of the sensor...and it would be a best compromise for a scene outside of the range of the sensor.
I don't see where any "reduction of exposure" to "limit the tones captured" does you any good...it all comes down to getting a proper capture and setting the proper tone responce curve...doesn't it?
I would ALWAYS prefer to take tones down in post than trying to lighten tones in post. It's easy to make your 1/4 tones darker but bloody hard to get them lighter without getting into noise.
I also don't have a problem tweaking highlights to increase texture and contrast. So even if my exposure to maintain textural highlights clumps a lot of data in the upper levels capturered, there are good techniques to bring those textures out. Careful tone curves in Camera Raw, dual processing in Camera Raw, local contrast adjustment in Photoshop, etc.
-
Wow. Seven pages since I last visited this topic. Pardon me if I don't read it all but considering what I've read on page 6 and 7, not much new ground has been treaded since page 2.
Maybe I'm missing something obvious. Or perhaps some others are missing the obvious...?
With regard to the comments by Jeff Schewe:
[...]
>And while you may think post processing is less desirable, I would argue that an artist will do whatever it takes to get the image to look the way they want.
I never said that post processing is less desireable, I suggested that I don't appreciate the over processed look of some of the expose-to-the-right imagery. I do post processing, layer adusts, masks and the like. I even did the Schewe edge sharpen routine until the new Photoshop sharpening showed up.
[...][a href=\"index.php?act=findpost&pid=63999\"][{POST_SNAPBACK}][/a]
How could you possibly know someone used ETTR? Over-processed is over-processed regardless of the use of ETTR.
While PS RAW histograms display separate channels, do in-camera histograms do so also? If not, won't ETTR relying on histograms result in potential clipped channels?
Newer cameras are coming out with separate RGB histograms but many others display luminance histograms. So, yes, ETTR can lead to clipped channels if you aren't careful to consider the subject matter you are photographing.
-
Rags' contention that less exposure may be better has it's roots in manufacturing a specific number of tones for a picture rendering. While zone V is drastic for some scenes and targets, it's not all that unbelievable for some capture devices. Why? Let's compare the number of tones in 12 bits (4096) with the signal-to-noise ratio of a high end camera back (4000:1). We've got approximately 4000 units in each comparative and the equation suggests that the noise in 12 bits is approximately 1 bit. In 12 stops of range, zone I has noise (0 is black) and the others are noise free.
For a normal scene (for example, let's say it's a 6 stop scene), one could expose right, place the highest important value way high with the histogram and no zone in the scene has noise. One could also expose the highest important on zone VII, and no zone in the scene has noise.
Richard Chang
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=63972\")
Your contention that your camera back, which I take to be the Megavision E5, has no noise above Zone I is not credible. That camera is based on a Kodak CCD, possibly the KAF 22000CE, which has 9 by 9 micron sensors. I do not have the specifications for that sensor, but for the following analysis I assume they are similar to the KAF-16801LE. If these specs are not representative for your sensor, please supply the correct specs, but the quoted specs should be close enough to demonstrate the principles.
[a href=\"http://www.kodak.com/global/en/digital/ccd/products/fullframe/KAF-16801E/specifications.jhtml?id=0.1.4.6.4.27.4&lc=en]http://www.kodak.com/global/en/digital/ccd....6.4.27.4&lc=en[/url]
All digital sensors have photon counting noise, which has a Poisson distribution and the standard deviation of this noise is equal to the square root of the number of photons collected. Another source of noise is the read noise. Dr. Roger Clark has done an analysis of noise for the Canon EOS 1D Mark II and the table shown is a similar analysis for the Kodak sensor, which has a full well capacity of 100,000 electrons and a read noise of 15 electrons at 1MHz.
http://www.clarkvision.com/imagedetail/eva...-1d2/index.html (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html)
For Zone 10 at base ISO, the sensor can utilize the full well electron count, and each of the other zones will have half the electrons of the preceeding as shown. For the darker zones, there must be considerable noise, better than the Canon because of the larger pixel size but only by a factor of 2 or so. If you shoot above base ISO, the full well will not be utilized and noise will be greater. In contrast to your statement that your camera is free of noise in Zone 10, the noise is actually greatest there, but because of the high S/N ration it is not evident.
[attachment=526:attachment]
In view of this analysis, I think that your shots could benefit from exposure to the right. If you disagree, please inform us why the laws of physics do not apply to your camera.
-
All digital sensors have photon counting noise, which has a Poisson distribution and the standard deviation of this noise is equal to the square root of the number of photons collected.
"Poisson distribution"? I don't know what the heck that is, but I sure love the implication...noise because of POISSON!!!
Yeah, yeah, I found out it's called that because of Siméon-Denis Poisson who did a probability theory and published Recherches sur la probabilité des jugements en matières criminelles et matière civile ("Research on the Probability of Judgments in Criminal and Civil Matters") in 1838. Interesting a theory about the Probability of Judgments in Criminal and Civil Matters has implications in 2006 digital capture...
Cool!
Course, I still don't understand it.
:~)
-
"Poisson distribution"? I don't know what the heck that is, but I sure love the implication...noise because of POISSON!!!
Yeah, yeah, I found out it's called that because of Siméon-Denis Poisson who did a probability theory and published Recherches sur la probabilité des jugements en matières criminelles et matière civile ("Research on the Probability of Judgments in Criminal and Civil Matters") in 1838. Interesting a theory about the Probability of Judgments in Criminal and Civil Matters has implications in 2006 digital capture...
Cool!
Course, I still don't understand it.
:~)
[a href=\"index.php?act=findpost&pid=64142\"][{POST_SNAPBACK}][/a]
Jeff,
Look at any statistics book on sampling distributions. The major ones are binomial, Poisson, and normal. If you read on in Wikipedia, you will see the Poisson distribution is also useful in assessing random events such as radioactive decay, road kill, and the number of soldiers killed by horse-kicks each year in each corps in the Prussian cavalry. It is not reassuring that a process for evaluating random events applies in court proceedings. In any event, the Poisson distribution also applies to photon sampling.
I can't tell if you are being sarcastic or not, but it is really interesting that all of these events can be evaluated by the Poisson process.
-
I can't tell if you are being sarcastic or not, but it is really interesting that all of these events can be evaluated by the Poisson process.
[a href=\"index.php?act=findpost&pid=64145\"][{POST_SNAPBACK}][/a]
Not sarcastic at all...I just think it's interesting that if one wants to, one can delve deep into the math of so many aspects of photography and digital imaging. And yes, I had to look it up on Wikipedia...course my eyes glazzed over cause I don't have the math :~)
My wife is constantly amazed by the fact that I can even use a computer to run Photoshop because I am so math challenged. But I get along...thankfully, I rely on Thomas Knoll and a real good interface between the math and the functions of Camera Raw/Photoshop!
-
Not sarcastic at all...I just think it's interesting that if one wants to, one can delve deep into the math of so many aspects of photography and digital imaging. And yes, I had to look it up on Wikipedia...course my eyes glazzed over cause I don't have the math :~)
My wife is constantly amazed by the fact that I can even use a computer to run Photoshop because I am so math challenged. But I get along...thankfully, I rely on Thomas Knoll and a real good interface between the math and the functions of Camera Raw/Photoshop!
[a href=\"index.php?act=findpost&pid=64147\"][{POST_SNAPBACK}][/a]
Yes, the math of the Poisson distribution can be a bit daunting when one has to calculate the mean, but fortunately one need not know the math to apply the concept. The nice thing about the Poisson distribution is that the mean and variance (square of the standard deviation) are the same. We measue the mean in Photoshop or stipulate it in the model and the square root function is available on our computers or calculators.
I am still waiting for Richard's response. I can't imagine how he can climb out of the hole into which he has dug, but it will be interesting. In any event, the math of noise justifies ETTR as much as does the number of levels, originally brought to our attention by Mr. Knoll's campfire side discussions with MR.
-
I haven't scanned any film since I went digital, so I have no experience recovering highlights in that situation and did not even know it was possible. Fortunately, you still have your original and can re do it.
The single histogram on most cameras is unfortunately a luminance histogram, not what Adobe calls an RGB histogram. In the luminance histogram, the RGB colors are weighted according to the sensitivity of the eye, and the red and blue channels do not get much weighting.
http://www.cambridgeincolour.com/tutorials/histograms2.htm (http://www.cambridgeincolour.com/tutorials/histograms2.htm)
With a lumonsity histogram it is very easy to blow a blue or red channel. Here is an example from my own experience. I exposed this red flower to the right according to the lumonsity histogram on my D70. In Photoshop, the lumonsity histogram looks fine (look at the black and white histogram) and resembles what I saw on the camera. However, the RGB histogram shows clipping of the red channel. My new D200 has separate RGB histograms which would avoid this problem, as do the professional Nikons and Canons. If a camera had only one histogram, one would think that it should be what Adobe calls the RGB histogram so as to avoid this problem.
[attachment=522:attachment]
[attachment=523:attachment]
BTW, the camera histogram does not show the RAW channels, but those of the JPEG preview, which has been converted to the assigned color space with the current camera settings. If you use ProPhotoRGB in ACR, the red clipping does not occur in the wider color space. Unfortunately, this space is not available on most cameras. (These screnshots are posted as GIF to better show text, hence the posterization in the colors)
[attachment=524:attachment]
[attachment=525:attachment]
[a href=\"index.php?act=findpost&pid=64089\"][{POST_SNAPBACK}][/a]
I'll save my response to recovering a scan's highlights for another thread. This one is getting rather lengthy. (But please don't stop.)
The referred article about histograms is very educational and well written, thanks. So ETTR is not as simple as just gauging by the histogram. One must also take into consideration a histogram's clippability (!?) and color space . More questions:
When is a camera's histogram available for viewing: before or after taking the shot?
If a camera's histogram can display separate rgb channels, will you ETTR so that no channel is clipped?
Of course, when in doubt, bracket, just like in the "good old" film days. The difference is burning storage rathen than film. How well do the higher end digital cameras support bracketing?
-
Sorry, Ray, but I don't get your point. [a href=\"index.php?act=findpost&pid=64098\"][{POST_SNAPBACK}][/a]
Well, what can I say, bjanes! I don't understand why you don't get my point.
Let me re-phrase it. If you unintentionally overexpose a blue sky, which you might well do when attempting to expose to the right, or even when simply using the camera's evaluative meter mode, there is a hue shift towards cyan which is not pleasant. This, I surmise, is due to the blue channel clipping more than the green channel.
If you unintentionally overexpose a grey, cloudy sky, or even a blue sky with clouds, the clouds usually retain their neutral greyness, although of course there is some loss of detail.
Now since I'm not in the habit of overexposing skies, I have trouble finding appropriate examples. However, on my way to the pub this afternoon, I stopped by the roadside to take a few shots just to confirm that I was not just imagining this effect.
[attachment=528:attachment]
-
Well, what can I say, bjanes! I don't understand why you don't get my point.
Let me re-phrase it. If you unintentionally overexpose a blue sky, which you might well do when attempting to expose to the right, or even when simply using the camera's evaluative meter mode, there is a hue shift towards cyan which is not pleasant. This, I surmise, is due to the blue channel clipping more than the green channel.
If you unintentionally overexpose a grey, cloudy sky, or even a blue sky with clouds, the clouds usually retain their neutral greyness, although of course there is some loss of detail.
Now since I'm not in the habit of overexposing skies, I have trouble finding appropriate examples. However, on my way to the pub this afternoon, I stopped by the roadside to take a few shots just to confirm that I was not just imagining this effect.
[attachment=528:attachment]
[a href=\"index.php?act=findpost&pid=64168\"][{POST_SNAPBACK}][/a]
With all due respect, Ray, the overexposure of the clear blue sky in your demonstration was hardly unintentional, and the fact that you could find no good example in your archives proves that it is not common in your experience. What did your meter, evaluative or spot, read for the overexposed sky and the rest of the scene? Would that degree of overexposure likely result from anything other than carelessness or deliberate attempt?
Since the sky is predominately blue, of course the blue channel will clip first. Minus 2 EV is beyond ACR's limits for highlight recovery. If you overexposed a red flower, the red channel would clip first. However, with a neutral scene, the channel that clips first is usually determined by the spectral response of the RGB channels. In the case I demonstrated, it is the green channel that clips first. So, again, what is your point?
-
So, again, what is your point?
Well now you've got me going back to page 6 to find out what the point of all this is. I think the point in general is that you surprised me by arguing with a common experience I had when I first started using the D60 with evaluative metering, several years ago.
You also made the following statement which doesn't accord with my experience.
It takes quite a bit of overexposure to blow out blue sky, since neither the luminance nor saturation of clear blue sky is that high. Much more likely, is blowing out of the clouds in a blue sky as shown in the shot of the Golden Gate Bridge in Bruce Fraser's article.
In the examples I've shown, both clouds and blue sky have partially blown with 2 stops overexposure. With one stop overexposure, which I would consider an ETTR situation for this shot, but which wasn't really necessary anyway, all detail has been recovered with -1 EC in ACR. Both clouds and sky seem to be equally susceptible to blowing, but the sky is more of a problem subjectively because of the hue shift.
-
Well now you've got me going back to page 6 to find out what the point of all this is. I think the point in general is that you surprised me by arguing with a common experience I had when I first started using the D60 with evaluative metering, several years ago.
You also made the following statement which doesn't accord with my experience.
In the examples I've shown, both clouds and blue sky have partially blown with 2 stops overexposure. With one stop overexposure, which I would consider an ETTR situation for this shot, but which wasn't really necessary anyway, all detail has been recovered with -1 EC in ACR. Both clouds and sky seem to be equally susceptible to blowing, but the sky is more of a problem subjectively because of the hue shift.
[a href=\"index.php?act=findpost&pid=64181\"][{POST_SNAPBACK}][/a]
Arguing with a common experience? If the experience is so common, why couldn't you recover an example in your archives? Why does Ansel Adams list clear northern sky as Zone V?
Since today is overcast and I don't have any examples of blown blue skies in my archives either, I did an experiment with a MacBeth color checker exposed under daylight (more or less, light entering my window on an overcast day) with proper exposure and with +2 EV exposure.
Here is the properly exposed target. Note the values for sky (3rd column, 1st row)
[attachment=529:attachment]
Here is the target with +2EV exposure:
[attachment=532:attachment]
Here is raw file with +2 EV as shown with DCRaw. Note that in the sky panel, no channels are blown and green exhibits the highest value.
[attachment=530:attachment]
Here is the +2EV exposure with -2EV in ACR. Since the sky patch had no blown channels, highlight recovery was possible for the sky, but other colors are clipped as shown in the histogram.
[attachment=531:attachment]
Again, it is not reasonable to expect to recover 2 stops of overexposure with ACR's highlight recovery. The example I gave with a Nikon camera may not apply to your Canon if it has different RGB filters. You should really look at your raw files without white balance and see what is going on. ETTR is NOT overexposure. It is exposure to the right of the histogram, not blowing highlights. Highlight recovery is for when you err in exposing to the right.
-
BJL,
Yes I have. It's probably a consequence of living in a sunny clime where brightness levels are greater than in northern Europe.
[a href=\"index.php?act=findpost&pid=63995\"][{POST_SNAPBACK}][/a]
I presume that by "brightness levels" you mean high "subject brightness range", or scenes of "high contrast", or in the Ansel Adam's jargon above, "long scale subjects". Because high overall brightness of a scene is not a problem in itself: just use sufficiently high shutter speeds to avoid over-exposure.
The issue I see a range of many stops from highlights to significant shadow details, or from an overall meter reading to far brighter highlights. In my experience those are also out of the range of the "straight" prints that I was asking about, into the territory needing tonal compression with low contrast settings, tone curve fiddling, or with film, printing with low contrast grades and perhaps developing for low contrast.
Anyway those sunny scenes are probably easily handled at base ISO speed like 100; the case I am worrying about is when base ISO speed will force one into underexposure due to shutter speed (and DOF) needs.
P. S. As an Australian who has lived and done much photography in the southern parts of the US, I know the problems of high subject brightness range!
-
I presume that by "brightness levels" you mean high "subject brightness range", or scenes of "high contrast", or in the Ansel Adam's jargon above, "long scale subjects". Because high overall brightness of a scene is not a problem in itself: just use sufficiently high shutter speeds to avoid over-exposure.
Yes. That's what I mean, high subject brightness range, which I believe is related to higher brightness levels in general because the parts of the scene in direct light will be brighter to a greater degree than the shadows will be, which rely upon indirect light.
The issue I see a range of many stops from highlights to significant shadow details, or from an overall meter reading to far brighter highlights. In my experience those are also out of the range of the "straight" prints that I was asking about, into the territory needing tonal compression with low contrast settings, tone curve fiddling, or with film, printing with low contrast grades and perhaps developing for low contrast.
I agree that such a wide DR will be out of the range of 'straight' prints, but aren't we all into some degree of digital processing of images before printing. I use the shadow/highlight tool in Photoshop a lot; also dual conversion. It's now possible to create a sort of composite image of parts from the one shot by isolating each significant part of the image in PS and processing that part as though it was a separate image.
Since most of my images are archived on DVD in a different location to where I am writing this, I can't drag out an image which would illustrate this well, but I have in mind a particular shot of a waterfall in a rainforest taken after a few rare weeks of good rainfall. The water was splashing over rocks near the top, creating an impressive spray of droplets which I wanted to capture (rather than the usual silky smooth flow that most people aim for). The sun was on part of the fall, but the foot of the fall, where there were interesting moss-covered boulders, pools and foliage, was in deep shadow. This was clearly a very high DR scene. I'd exposed for the highlights at ISO 100 and captured the spray clearly and sharply without blowing highlights significantly, but failed to get a decent rendering in post processing of the rocks in the shade. Neither Neat Image nor dual conversion was of much help. The image was therefore spoiled in my view; an experiment which didn't work because the Camera (the D60) had insufficient DR.
-
>Your contention that your camera back, which I take to be the Megavision E5, has no noise above Zone I is not credible.
You are right, and according to the included table, I stand corrected.
> In contrast to your statement that your camera is free of noise in Zone 10, the noise is actually greatest there, but because of the high S/N ration it is not evident.
Not evident is a critical concept. Evident by what means? I took a look at some of my images, specifically just brighter than zone I. You are correct and because of your post, I went looking for noise and I did find a slight structure which became evident at 400% magnification. It does not render on a print, even at 24x32", but it is there. If I don't see the noise render, must I avoid the advantages I see in highlight tonality, just to remove an artifact that can't be seen?
>In view of this analysis, I think that your shots could benefit from exposure to the right.
I think the analysis, at least when based on the performance of the image when rendered on the output target, would suggest that I can choose the more important attribute as I see it; remove some noise that I cannot see on the print, or take the advantage of increased flexibility in crafting highlight detail upwards to the paper's highest separable value. This careful examination of the noise you've alerted me to, has allowed me to eval the danger, and subsequently decide that it's not worth worrying about.
If you disagree, please inform us why the laws of physics do not apply to your camera.
I do agree with the notion of ETTR, for lower s/n capture devices. The laws of physics do apply to our camera and you are correct to steer me back from the error of my claims. I'll include your observation of the Kodak noise performance in my teaching, and I'll give you credit for alerting me to the perils of low valued noise.
Richard Chang
-
> In contrast to your statement that your camera is free of noise in Zone 10, the noise is actually greatest there, but because of the high S/N ration it is not evident.
Not evident is a critical concept. Evident by what means? I took a look at some of my images, specifically just brighter than zone I. You are correct and because of your post, I went looking for noise and I did find a slight structure which became evident at 400% magnification. It does not render on a print, even at 24x32", but it is there. If I don't see the noise render, must I avoid the advantages I see in highlight tonality, just to remove an artifact that can't be seen?
Richard Chang
[a href=\"index.php?act=findpost&pid=64251\"][{POST_SNAPBACK}][/a]
Richard,
Thank you for a gracious reply. The signal to noise ratio of a device capturing 100,000 electrons is absurdly high and far beyond 4 by 5 Velvia film. By not evident, I meant to the unaided human eye, which is the final arbiter of our photographic efforts. However, the digital S/N is progressively lower in the deep shadows and it is here that ETTR could help visually in Zone I as you noted. IMHO, use of base ISO can be more critical than ETTR for improved N/R.
-
When is a camera's histogram available for viewing: before or after taking the shot?
After taking the shot, since it's showing the distribution of the number of pixels according to luminance (or red, green and blue channels) in the picture you just took (or to be specific, in the JPEG preview).
If a camera's histogram can display separate rgb channels, will you ETTR so that no channel is clipped?
If you don't want clipping, yes.
Of course, when in doubt, bracket, just like in the "good old" film days.
I think bracketing is overrated, since it requires that nothing is in motion.
Mountains and wind-free days are your friends.
The difference is burning storage rathen than film. How well do the higher end digital cameras support bracketing?
Canon isn't very good at automatic exposure bracketing, offering only +/- 3 EV in three steps. Recent Nikon cameras offer +/- 5 EV, if I recall correctly. But bracketing manually isn't that hard, if the scene allows you the time to do it.
-
After taking the shot, since it's showing the distribution of the number of pixels according to luminance (or red, green and blue channels) in the picture you just took (or to be specific, in the JPEG preview).
If you don't want clipping, yes.
I think bracketing is overrated, since it requires that nothing is in motion.
Mountains and wind-free days are your friends.
Canon isn't very good at automatic exposure bracketing, offering only +/- 3 EV in three steps. Recent Nikon cameras offer +/- 5 EV, if I recall correctly. But bracketing manually isn't that hard, if the scene allows you the time to do it.
[a href=\"index.php?act=findpost&pid=64263\"][{POST_SNAPBACK}][/a]
Some metering methods, such as spot metering, bracketing (a digital camera's long lag time between raw shots is a killer), etc., do not work well if the subjects within an image are not stationary, or if the lighting is constantly changing. ETTR is also in this category if the histogram is displayed after a shot.
-
Some metering methods, such as spot metering, bracketing (a digital camera's long lag time between raw shots is a killer), etc., do not work well if the subjects within an image are not stationary, or if the lighting is constantly changing. ETTR is also in this category if the histogram is displayed after a shot.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=64465\")
Most cameras have a buffer that stores the shots while they are being processed and being written to the memory card. For example, in his tests of the Nikon D200 in DPReview, Phil noted that the camera could take 22 shots before slowing down. How much bracketing do you do?
[a href=\"http://www.dpreview.com/reviews/nikond200/page12.asp]http://www.dpreview.com/reviews/nikond200/page12.asp[/url]
-
Most cameras have a buffer that stores the shots while they are being processed and being written to the memory card. For example, in his tests of the Nikon D200 in DPReview, Phil noted that the camera could take 22 shots before slowing down. How much bracketing do you do?
http://www.dpreview.com/reviews/nikond200/page12.asp (http://www.dpreview.com/reviews/nikond200/page12.asp)
[a href=\"index.php?act=findpost&pid=64476\"][{POST_SNAPBACK}][/a]
The D200 numbers are indeed impressive.
Even though my film camera supports up to ~15 auto bracketed frames with one click (two clicks per 36 exposure roll!), I have never used it. I typically spot meter in manual mode, and can't use auto bracketing since it is not supported. In fact after getting used to metering accurately this way, I now only bracket in really challenging situations.
-
A few thoughts on this thread:
If your camera has one, always use an RGB histogram when shooting ETTR, to avoid clipping individual channels. You will also need to set a manual white balance such that equal RAW values result in equal RGB values when your camera generates the internal JPEG conversion used to generate the histogram. With Canon DSLRs, the daylight WB preset is pretty close, other brands and models will likely vary. This is critical when shooting in situations with lighting that differs significantly from daylight. Gelled incandescent stage lighting will fool you into blowing the red channel severely if you use auto WB, and shooting before dawn and after sunset will often blow the blue channel.
The whole point of ETTR is to maximize the dynamic range of the scene that is captured by the sensor, which will result in the greatest S/N ratio in the captured image data. With a high-DR sensor and a low-DR subject, you have some fudge room to work with, but even in such cases, if you do not practice ETTR you are still compromising your image to some extent. Whether this compromise materially affects the final print will vary greatly from image to image depending on many factors, but I see no reason not to strive for the ideal exposure (non-specular highlights within 1/2 stop of clipping, but not clipped) that will allow the greatest flexibility in post-processing and printing.
We don't shoot JPEGs for most serious work, even though JPEGs are "good enough" in most cases; we shoot RAW to take advantage of the increased control and processing flexibility advantages the RAW format offers over JPEG. Few would argue that shooting RAW constitutes a license to "shoot sloppy" even though RAW files tend to be far more salvageable when Murphy's Law strikes. In the same way, properly applying ETTR gives one a margin of flexibility that may not be absolutely needed all the time, but sometimes can really save one's backside, especially when shooting difficult high-DR subjects. There are times one can get away with being sloppy and deviate from the ideal without negatively affecting the final print enough to be noticed, but that's no excuse to be deliberately sloppy about exposure, or any other part of our work.
Canon 1-series cameras can shoot at least 7 auto-bracketed frames 3 stops apart. Consult the manual regarding Personal Function 8 to set this up. I have my 1Ds and 1D-MkII set for 5-frame brackets when PFn 8 is on. For DR blending, 5 frames shot at 1-2 stop intervals is sufficient for just about any real-world subject.
-
...if you do not practice ETTR you are still compromising your image to some extent. Whether this compromise materially affects the final print will vary greatly from image to image depending on many factors, but I see no reason not to strive for the ideal exposure (non-specular highlights within 1/2 stop of clipping, but not clipped) that will allow the greatest flexibility in post-processing and printing.
[a href=\"index.php?act=findpost&pid=64686\"][{POST_SNAPBACK}][/a]
That's a very reasonable position to take, Jonathan . My only reservation would be in situations where one is striving to capture the moment and one either has to make a rapid decision regarding exposure, or preset a combination of ISO, aperture and EC for use in evaluative metering mode, that one judges will cover most eventuallities for the conditions.
I would rather sacrifice ETTR considerations for the sake of getting the shot and I would rather expose one stop under the ETTR threshold than one stop over.
However, for me, the situation has now been complicated with ownership of a 5D. With the D60, a shot that was 2 stops underexposed at ISO 100 was hardly more noisy than an ETTR shot at ISO 400. I believe a similar situation applies to the Nikon D70 and D200.
However, with this extraordinary 'noise reduction' that's taking place at high ISOs with the 5D, it now really does seem to be sloppy not to strive for ETTR in every situation. I shall buy some 4gb, or even 8gb cards for my next trip and autobracket every situation where I don't have time to go through the spot metering routine, and those occasions are frequent. I just hope I don't get overwhelmed with images to sort through on my return.
-
A few thoughts on this thread:
If your camera has one, always use an RGB histogram when shooting ETTR, to avoid clipping individual channels. You will also need to set a manual white balance such that equal RAW values result in equal RGB values when your camera generates the internal JPEG conversion used to generate the histogram. With Canon DSLRs, the daylight WB preset is pretty close, other brands and models will likely vary. This is critical when shooting in situations with lighting that differs significantly from daylight. Gelled incandescent stage lighting will fool you into blowing the red channel severely if you use auto WB, and shooting before dawn and after sunset will often blow the blue channel.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=64686\")
The idea of setting a white balance such that the camera histogram shows the raw channels with gamma correction and tone curve applied is a good one, but daylight balance will not accomplish the task for either the Canon EOS 1D Mark II nor the Nikon D200.
For daylight exposure of a white target such that the RGB values in an ACR conversion are equal and about 240 in sRGB, the Canon EOS 1D Mark II and the Nikon D200 produce similar values in the RAW channels. I have a CR2 file exposed in daylight and white balanced so that the RGB channels read 236 in the conversion. The RAW values as determined by DCRaw are 77, 148 and 115. For the D200 NEF the sRGB values are 241 and the RGB channels are 95, 178, and 136. The Canon allows a bit more head room, but the ratios of the channels are very similar. Both raw files have a decidedly green cast when viewed without white balance.
To white balance the above raw files, one multiplies the raw red value by 1.85 and the raw blue value by 1.26.
To have the channels of the converted RGB file equal to those of the raw file under the above conditions, one would have to apply a reverse white balance by multiplying the red channel by 1/1.85 and the blue channel by 1/1.26.
Julia Borg has published white balance coefficients for Nikon cameras showing similar with multipliers for the D200. Unfortunately, there is no color temperature having red and blue multipliers of 1.0 and 1.0.
[a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm[/url]
Iliah Borg has devised a uniwhite balance that can be loaded into Nikon cameras so that the histograms represent the actual values in the raw channels after application of gamma correction and the tone curve. Rather than using a color temperature, the white balance specifies the multiplers directly (red 0.55 and blue 0.74). I don't how this would be done for the Canon.
Shown below is the DCRaw conversion of the Canon raw file with no white balance. Since the file is linear (gamma 1.0), it appears dark.
[attachment=545:attachment]
-
To have the channels of the converted RGB file equal to those of the raw file under the above conditions, one would have to apply a reverse white balance by multiplying the red channel by 1/1.85 and the blue channel by 1/1.26.
Julia Borg has published white balance coefficients for Nikon cameras showing similar with multipliers for the D200. Unfortunately, there is no color temperature having red and blue multipliers of 1.0 and 1.0.
http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm (http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm)
Iliah Borg has devised a uniwhite balance that can be loaded into Nikon cameras so that the histograms represent the actual values in the raw channels after application of gamma correction and the tone curve. Rather than using a color temperature, the white balance specifies the multiplers directly (red 0.55 and blue 0.74). I don't how this would be done for the Canon.
[a href=\"index.php?act=findpost&pid=64746\"][{POST_SNAPBACK}][/a]
A note to clairify the uniwb of Mr. Borg. It sets the red and blue coefficients for white balance conversion to 1.0 and 1.0. The reverse WB is necessary only if one is going from an image that is already white balanced back to what it would have been in RAW.
To verify the behavior of uniwb, I took a photograph of a Macbeth color checker in daylight with the D200 with the uniwb as a preset for WB and opened the NEF in ACR. The preview with the as shot WB looked very green as expected. ACR reported the as shot WB as 5000 tint +10. To white balance, setting ACR to daylight did not work, and the picture still appeared quite green. White balance by clicking with the WB eyedropper of ACR on C4R2 was successful and the WB in ACR read 6750 tint +134. This was not the behavior that I expected.
In Nikon Capture, the uniwb shot showed normal WB when using the daylight setting rather than the as shot. Entering the multipliers manually as 1.82 and 1.38 for the red and blue channels also worked.
-
White balance by clicking with the WB eyedropper of ACR on C4R2 was successful and the WB in ACR read 6750 tint +134. This was not the behavior that I expected.
[a href=\"index.php?act=findpost&pid=64775\"][{POST_SNAPBACK}][/a]
bjanes,
It sounds to me that in the processing of reducing one problem (dynamic range) you are increasing another problem (accurate color). I don't think I would like to rely upon the eyedropper to get an accurate WB.
-
bjanes,
It sounds to me that in the processing of reducing one problem (dynamic range) you are increasing another problem (accurate color). I don't think I would like to rely upon the eyedropper to get an accurate WB.
[a href=\"index.php?act=findpost&pid=64855\"][{POST_SNAPBACK}][/a]
Use of the eyedropper on a neutral target of relatively high luminance is the best WB. Alternatively, one can do a preset WB from a digital gray card. I don't think your intuition without any testing is a good basis on which to base conclusions.
-
Use of the eyedropper on a neutral target of relatively high luminance is the best WB.
Maybe! But do all images contain such a neutral target of high luminance?
Alternatively, one can do a preset WB from a digital gray card
Doesn't doing that get you back to the initial position of compromising DR?
I don't think your intuition without any testing is a good basis on which to base conclusions.
I haven't arrived at any conclusions on this technique you've described. I'm just expressing an imression that the procedure sounds difficult, fiddly and time-consuming.
Could you post a couple of color corrected, WB corrected, sample real-world images demonstrating the increase in DR (or reduction in noise) that's possible with this technique?
-
Maybe! But do all images contain such a neutral target of high luminance?
Doesn't doing that get you back to the initial position of compromising DR?
I haven't arrived at any conclusions on this technique you've described. I'm just expressing an imression that the procedure sounds difficult, fiddly and time-consuming.
Could you post a couple of color corrected, WB corrected, sample real-world images demonstrating the increase in DR (or reduction in noise) that's possible with this technique?
[a href=\"index.php?act=findpost&pid=64890\"][{POST_SNAPBACK}][/a]
Including a neutral target is a standard method for more advanced photographers. You don't have to include it in every picture, but only when shooting conditions change. You can't use uniwb (multipliers = 1) and the preset at the same time obviously, but you can store the preset for later use and use the uniwb for a better idea of the contents of the RGB channels on the histogram while you are shooting and then apply the proper WB in post. Since the WB is only a tag and does not affect the image data, it has no effect on dynamic range.
Balancing the channels can gain up to 3/4 stop in dynamic range and this has been demonstrated recently on the DPReview Nikon D2 forum with high resolution shots. It may not be worth the trouble in most instances, and I don't normally find need for it, but I was just following up on a suggestion by Jonathin and I think it is a very interesting concept. If you want to save time, use auto WB or just get a P&S.
-
To verify the behavior of uniwb, I took a photograph of a Macbeth color checker in daylight with the D200 with the uniwb as a preset for WB and opened the NEF in ACR. The preview with the as shot WB looked very green as expected. ACR reported the as shot WB as 5000 tint +10. To white balance, setting ACR to daylight did not work, and the picture still appeared quite green. White balance by clicking with the WB eyedropper of ACR on C4R2 was successful and the WB in ACR read 6750 tint +134. This was not the behavior that I expected.
In Nikon Capture, the uniwb shot showed normal WB when using the daylight setting rather than the as shot. Entering the multipliers manually as 1.82 and 1.38 for the red and blue channels also worked.
[a href=\"index.php?act=findpost&pid=64775\"][{POST_SNAPBACK}][/a]
Is the ACR behaviour due to the fact that "as shot" is not really as shot, but ACR's best guess as to what "as shot" really was, due to the infamous WB "encryption"?
Karsten
-
Is the ACR behaviour due to the fact that "as shot" is not really as shot, but ACR's best guess as to what "as shot" really was, due to the infamous WB "encryption"?
Karsten
[a href=\"index.php?act=findpost&pid=65497\"][{POST_SNAPBACK}][/a]
No, the ACR rendering as shot is similar to the Nikon Capture as shot, so I conclude that ACR is reading the recorded white balance correctly. However, selection of daylight WB in ACR produces little change in the rendering, which is still far too green. In Nikon Capture, the daylight setting produces the intended correct WB. When I set the WB with the ACR eyedropper, the resultant color temperature is 6750 +135
-
Is the ACR behaviour due to the fact that "as shot" is not really as shot, but ACR's best guess as to what "as shot" really was, due to the infamous WB "encryption"?
Karsten
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=65497\")
The answer to this ACR behavior was given by Thomas Knoll (ACR and Photoshop creator) on the Adobe Camera Raw forum:
[a href=\"http://www.adobeforums.com/cgi-bin/webx/.3bb6a85c.3bbfce73/2]http://www.adobeforums.com/cgi-bin/webx/.3bb6a85c.3bbfce73/2[/url]
The Nikon cameras have a doubble exposure feature, which sets the white balance coefficients to 1.0. With the white balance coefficients set to 1.0, ACR assumes that it is dealing with a double exposure situation and uses different processing.
-
Nothing wanting to go too OT, I was wondering if this new target would help in a number of ways:
http://www.babelcolor.com/main_level/White_Target.htm (http://www.babelcolor.com/main_level/White_Target.htm)
I just got one and will play. But right off the bat, I could see it might be ideal for bracketing for deciding where the exposure for the best "expose to the right" values should be as well as for setting neutrals. Thoughts?
-
Nothing wanting to go too OT, I was wondering if this new target would help in a number of ways:
http://www.babelcolor.com/main_level/White_Target.htm (http://www.babelcolor.com/main_level/White_Target.htm)
I just got one and will play. But right off the bat, I could see it might be ideal for bracketing for deciding where the exposure for the best "expose to the right" values should be as well as for setting neutrals. Thoughts?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=65596\")
$69 for a little piece of plastic seems a bit high to me. The WhiBal for $29.99 seems like a better deal to me, but I haven't used either one of these devices. Thom Hogan (Nikon guru) does not recommend white for white balance and says gray works better, even though Nikon says white or gray works. I do not know what Canon suggests.
[a href=\"http://www.rawworkflow.com/products/whibal/index.html]http://www.rawworkflow.com/products/whibal/index.html[/url]
As for ETTR, why look for the small white spike in the histogram? --it wouldn't even be visible at a distance. As discussed previously, the histogram may indicate clipping when there is none and using it for exposure may result in underexposure. Why not just meter from a highlight?
-
$69 for a little piece of plastic seems a bit high to me. The WhiBal for $29.99 seems like a better deal to me, but I haven't used either one of these devices.
Well not all little pieces of plastic are created equally! The spectral qualities of this new target might well be worth the extra bucks. I can (and will) measure it with my Spectrophotometer. I don't have the $29.00 goodie so I can't compare it but with the $69 goodie, the spec's are impressive. Again, whatever the plastic is and how it reads spectrally is pretty important here. If not, we could use any little piece of white plastic.
-
As for ETTR, why look for the small white spike in the histogram? --it wouldn't even be visible at a distance. As discussed previously, the histogram may indicate clipping when there is none and using it for exposure may result in underexposure. Why not just meter from a highlight?
[a href=\"index.php?act=findpost&pid=65605\"][{POST_SNAPBACK}][/a]
I've found that Dale Cotton's suggestion for a reasonably accurate ETTR exposure works quite well. You might have come across it in a previous thread. Essentially, use the spot meter on the brightest part of the image, which could be a fluffy white cloud in a landscape, or a white paper napkin on a table indoors. Use a shutter speed 3 stops slower than the spot meter reading. Voila!
-
>The Gray Cards within the WhiBal™ system have both *a and *b Lab channel values of less than 1 >at D50 and are spectrally "flat".
According to the spec's for the BabeColor white it's a* and b* are -0.08/-0.04 so if we buy into the spec's, it's a significantly better piece of plastic. The actual spectral plot from 400nm to 715nm are provided (and while it's pretty flat, it's not perfectly flat as you'd expect). So the spec's given seem to be more "real world" and not as ambiguous as the other product. Further it says it's thermally stable and water proof.
-
>The Gray Cards within the WhiBal™ system have both *a and *b Lab channel values of less than 1 >at D50 and are spectrally "flat".
According to the spec's for the BabeColor white it's a* and b* are -0.08/-0.04 so if we buy into the spec's, it's a significantly better piece of plastic. The actual spectral plot from 400nm to 715nm are provided (and while it's pretty flat, it's not perfectly flat as you'd expect). So the spec's given seem to be more "real world" and not as ambiguous as the other product. Further it says it's thermally stable and water proof.
[a href=\"index.php?act=findpost&pid=65609\"][{POST_SNAPBACK}][/a]
A possible limitation of the BabeColor for white balance is that it has a reflectivity of 99% and it likely is the object with the highest luminance in the shot, excluding specular highlights. If the color channels are slightly blown in the target, ACR will not permit a white balance to be taken. That is one reason why a light gray target is preferable for white balance.
-
A possible limitation of the BabeColor for white balance is that it has a reflectivity of 99% and it likely is the object with the highest luminance in the shot, excluding specular highlights. If the color channels are slightly blown in the target, ACR will not permit a white balance to be taken. That is one reason why a light gray target is preferable for white balance.
[a href=\"index.php?act=findpost&pid=65690\"][{POST_SNAPBACK}][/a]
Not sure. Here's the test I just did (it may be valid, may not; I'm grasping at all this too).
Shot a Macbeth Color Checker (24 patch) with the Babelcolor target in full sun (clear blue Santa Fe sky). My Minolta Flash meter set for incident light gave me a perfect F8 @ 250th at ISO 100. I set the Canon Rebel XT to manual and bracketed (F8 normal then F7.3 and F6.1 for over and F9 and F10 for under exposure. This is the smallest manual bracket in F stops the camera allows).
Brought all the RAW files into ACR in film strip mode, turned off all the Auto settings. F8 (the "normal" exposure) does seem to be the best as the White tile reads 250/251/252. Any other bracket either blows it out to 255 or is under exposed with respect to white. This leads me to believe that the ISO of the camera is correct based on using this external meter.
The white isn't white (but it's close). Note that the Macbeth White in the "correct" exposure reads 246/247/248. So it is neutral but not as white as the white Babel target. Using ACRs White Balance eyedropper on the Bable target produces a value of 252/252/252 (the instructions suggest that 253/254/254 is expected but I assume this is "close enough"). For grins I tried to up the exposure to get another value out of this white so I could save all this as a new ACR default. Setting the exposure to 0.1 did this. Color seems to be a bit better with the white balance on this white square. All the gray squares in the Macbeth read neutral within 2 values of all RGB numbers.
As for other colors, well they might need work (like using the ACR script). The calibrate tab would help with reds and such. But it appears with the Auto settings off, the highlights are pretty darn close to where they need to be just below clipping IF (big if?) we assume that this white tile is where it should be based on the LAB values I got off of it and the 99% reflectance.
Long story short, I don't see that simply using it to white balance is an issue. Whites, blacks and grays are neutral numerically. Did using this as my exposure guide produce optimal EFTR? Can I use this for any other images?
-
Not sure. Here's the test I just did (it may be valid, may not; I'm grasping at all this too).
Shot a Macbeth Color Checker (24 patch) with the Babelcolor target in full sun (clear blue Santa Fe sky). My Minolta Flash meter set for incident light gave me a perfect F8 @ 250th at ISO 100. I set the Canon Rebel XT to manual and bracketed (F8 normal then F7.3 and F6.1 for over and F9 and F10 for under exposure. This is the smallest manual bracket in F stops the camera allows).
Brought all the RAW files into ACR in film strip mode, turned off all the Auto settings. F8 (the "normal" exposure) does seem to be the best as the White tile reads 250/251/252. Any other bracket either blows it out to 255 or is under exposed with respect to white. This leads me to believe that the ISO of the camera is correct based on using this external meter.
The white isn't white (but it's close). Note that the Macbeth White in the "correct" exposure reads 246/247/248. So it is neutral but not as white as the white Babel target. Using ACRs White Balance eyedropper on the Bable target produces a value of 252/252/252 (the instructions suggest that 253/254/254 is expected but I assume this is "close enough"). For grins I tried to up the exposure to get another value out of this white so I could save all this as a new ACR default. Setting the exposure to 0.1 did this. Color seems to be a bit better with the white balance on this white square. All the gray squares in the Macbeth read neutral within 2 values of all RGB numbers.
As for other colors, well they might need work (like using the ACR script). The calibrate tab would help with reds and such. But it appears with the Auto settings off, the highlights are pretty darn close to where they need to be just below clipping IF (big if?) we assume that this white tile is where it should be based on the LAB values I got off of it and the 99% reflectance.
Long story short, I don't see that simply using it to white balance is an issue. Whites, blacks and grays are neutral numerically. Did using this as my exposure guide produce optimal EFTR? Can I use this for any other images?
[a href=\"index.php?act=findpost&pid=65698\"][{POST_SNAPBACK}][/a]
For white balance with the Macbeth CC, Bruce Fraser and others suggest using the second patch in row 4 (OD 0.23, reflectance 59%, pixel value 202 with gamma 2.2). Patch 1 (OD 0.05, reflectance 89%, pixel value 242) also works but I think the rationale of patch 2 is to have better WB in the darker tones in case of non-linearity. The BableColor has 99% reflectance, OD 0.004, and pixel value 244 with gamma 2.2. A bit of overexposure would saturate the color channels and white balance would not be accurate and ACR would not allow it. Why not use a more conservative gray value?
Bruce is your buddy and business partner--why don't you get his input and tell us why he prefers R4C2 in the CC?
-
Brought all the RAW files into ACR in film strip mode, turned off all the Auto settings. F8 (the "normal" exposure) does seem to be the best as the White tile reads 250/251/252. Any other bracket either blows it out to 255 or is under exposed with respect to white. This leads me to believe that the ISO of the camera is correct based on using this external meter.
The numerical value for a properly exposed white item should read between 238 to 242. If you are getting 250, I'd guess off the top of my head you are about a half-stop over.
Long story short, I don't see that simply using it to white balance is an issue. Whites, blacks and grays are neutral numerically. Did using this as my exposure guide produce optimal EFTR? Can I use this for any other image
Yes, as you mentioned more exposure in this case resulted in clipping you obtained optimal ETTR for your camera and shooting conditions. (I'm assuming EFTR is "expose for the right") And yes, this can be used for any other image you have taken under the same light. If the exposure and light intensity remained the same (as in no fluctuating flash output cloud cover, etc), you can use that one frame to set exposure correction for the others as well.
Some thoughts on the Babel vs Whibal:
If absolute accuracy is needed in controlled situations, then the Babel I'm sure would be worth the extra cost. However, in most situations the WhiBal works jim-dandy and provides very good White-balance results. The bigger problems for most usage in my opinion is making sure you use it right and keep it clean. If you aren't careful, color casted light can reflect off the card and have an adverse effect on the results (sometimes quite severely). This should go without saying, but I've gotten quite a few files from clients where the card results were worthless since they had the thing tilted wrong.
One thing I like about the WhiBal is that it is neutral throughout the plastic which makes it quite durable. If the thing gets dirty or stained I can scrub it clean or even sand it down with fine-grit sand paper to restore it to like-new operating conditions. I didn't see any info on the Babel site indicating a similar feature.
-
The numerical value for a properly exposed white item should read between 238 to 242. If you are getting 250, I'd guess off the top of my head you are about a half-stop over.
Why? I thought the idea for EFTR was to get as close to blowing out white as possible without doing so. The L value of the target used is pretty close to L99 (I measured it with my EyeOne). I could see not wanting the Macbeth white to be there and instead around 240 but this white tile is about as white an object I'll ever encounter no?
Some thoughts on the Babel vs Whibal:
If absolute accuracy is needed in controlled situations, then the Babel I'm sure would be worth the extra cost. However, in most situations the WhiBal works jim-dandy and provides very good White-balance results. The bigger problems for most usage in my opinion is making sure you use it right and keep it clean. If you aren't careful, color casted light can reflect off the card and have an adverse effect on the results (sometimes quite severely). This should go without saying, but I've gotten quite a few files from clients where the card results were worthless since they had the thing tilted wrong.
They do tell you to keep it clean and NOT to scratch it.
So lets forget the white tile as a device to set white balance (although it seems to work fine) and instead use the above logic to say that using my external meter set to ISO 100 to get an exposure does produce a white that's close to ideal. Now the question is, how do (or can I) correlate this to the internal meter? My thought is it's seeing the world as 18% gray. In theory, if I placed a gray card in the same scene and got the same exposure (F8@250th) then my assumption would be the in camera meter is "correct" for EFTR. If it's not, I would need to use some exposure compensation to produce an in camera exposure that syncs up with the Minonlta. Does that sound reasonable?
Again, this is built on the assumption that exposing the white tile based on my Minolta produced the best exposure of the bracket series to be just shy of white clipping. At least using ACR (my preferred converter) with all Auto settings off.
-
Why? I thought the idea for EFTR was to get as close to blowing out white as possible without doing so.
Correct, but that doesn't lead to correct exposure, just optimum capture information. Exposure is optimized after the fact in the raw converter.
The L value of the target used is pretty close to L99 (I measured it with my EyeOne). I could see not wanting the Macbeth white to be there and instead around 240 but this white tile is about as white an object I'll ever encounter no?
Sometimes you may encounter whiter than white such a something with a slightly stronger amount of light shining on it that still needs to maintain some detail.
And that leads to one thing that needs consideration; highlight detail. If 250 were the optimum value for a properly exposed white object then that means you only have 5 values of tonal difference between that white object, something glowing white or a specular highlight for example. That is not much for the eye to discern between. 238-242 is the range for optimum whites while maintaining visible detail in those whites not to mention a natural appearance.
They do tell you to keep it clean and NOT to scratch it.
Sure, keeping it clean is something to strive for but when out and about things can get dirtied. I bought the 1st generation WhiBal when it first came out and at the time they were advertising its durability in stating that scratches would not effect readings since the plastic is neutral throughout. Based off this information, while not recommended by them I'm sure, the WhiBal can be recovered if the worst was to happen.
So lets forget the white tile as a device to set white balance (although it seems to work fine) and instead use the above logic to say that using my external meter set to ISO 100 to get an exposure does produce a white that's close to ideal. Now the question is, how do (or can I) correlate this to the internal meter? My thought is it's seeing the world as 18% gray. In theory, if I placed a gray card in the same scene and got the same exposure (F8@250th) then my assumption would be the in camera meter is "correct" for EFTR. If it's not, I would need to use some exposure compensation to produce an in camera exposure that syncs up with the Minonlta. Does that sound reasonable?
The White tile certainly can be used for WB but over time I have discovered the light grey square does provide better results more often. The difference is usually minimal and the times the white does not work properly is typically when there is a channel clipping or it is very close to channel clipping.
As to syncing the camera and meter, correct. This will give you proper exposure, or as proper as possible given the mechanical limitations of the equipment. This will not however set you up for using expose to the right as that involves over exposure. How much over-exposure depends entirely on the scene and what information you want to keep and throw away.
-
Correct, but that doesn't lead to correct exposure, just optimum capture information. Exposure is optimized after the fact in the raw converter.
This will not however set you up for using expose to the right as that involves over exposure. How much over-exposure depends entirely on the scene and what information you want to keep and throw away.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=65725\")
Dynamic,
I think that you are being a bit dogmatic and overbearing. As you should know, Andrew is an expert in digital imaging and color management and his views should not be dismissed lightly.
You seem to be using the definition of overexposure for negative film. With digital, the situation is somewhat different as summarized by Bruce Fraser in another post on the Adobe Forums involving exposure to the right:
"For the purposes of this discussion, 'overexposure' means blowing highlights you didn't want to blow, and "correct exposure" means holding exactly the highlight detail you wanted to keep. In terms of what the camera meter tells you, that may well translate to systematic overexposure, but that's because the meter isn't well-suited to digital because it's calibrated to 12% reflectance, which is all the way down at the 12% level in a linear capture."
[a href=\"http://www.adobeforums.com/cgi-bin/webx?14@@.3bbb579c/65]http://www.adobeforums.com/cgi-bin/webx?14@@.3bbb579c/65[/url]
One could define proper exposure as that exposure that optimizes capture information. For Ansel Adams and negative film, that involved exposing for the shadows. For digital capture, it involves exposing for the highlights. The nominal meter reading of the overall scene does not define proper expsure.
-
I've found that Dale Cotton's suggestion for a reasonably accurate ETTR exposure works quite well. You might have come across it in a previous thread. Essentially, use the spot meter on the brightest part of the image, which could be a fluffy white cloud in a landscape, or a white paper napkin on a table indoors. Use a shutter speed 3 stops slower than the spot meter reading. Voila!
[a href=\"index.php?act=findpost&pid=65608\"][{POST_SNAPBACK}][/a]
As mentioned in a few other threads, this is the same technique used to expose chromes in a film camera to preserve the highlight details. Nothing has changed except for the introduction of a new acronym ETTR.
-
Dynamic,
I think that you are being a bit dogmatic and overbearing. As you should know, Andrew is an expert in digital imaging and color management and his views should not be dismissed lightly.
Thanks however, I'm still trying to ensure that all this expose to the right, use the white tile stuff is working and makes scenes to me and others. I'm in the same boat with the rest of you <g>.
If there were a way to photograph an object and objectively evaluate that I was placing the most bits in the last stop (and I'd have to recognize real data from noise), that would be useful.
-
As mentioned in a few other threads, this is the same technique used to expose chromes in a film camera to preserve the highlight details. Nothing has changed except for the introduction of a new acronym ETTR.
[a href=\"index.php?act=findpost&pid=65783\"][{POST_SNAPBACK}][/a]
ETTR is similar to the exposure of chromes, but with chromes one exposes so that the tone placement is where it is desired in the final image. In ETTR, the image might be too light without exposure adjustment in the raw converter.
-
Dynamic,
I think that you are being a bit dogmatic and overbearing. As you should know, Andrew is an expert in digital imaging and color management and his views should not be dismissed lightly.
You seem to be using the definition of overexposure for negative film. With digital, the situation is somewhat different as summarized by Bruce Fraser in another post on the Adobe Forums involving exposure to the right:
"For the purposes of this discussion, 'overexposure' means blowing highlights you didn't want to blow, and "correct exposure" means holding exactly the highlight detail you wanted to keep. In terms of what the camera meter tells you, that may well translate to systematic overexposure, but that's because the meter isn't well-suited to digital because it's calibrated to 12% reflectance, which is all the way down at the 12% level in a linear capture."
http://www.adobeforums.com/cgi-bin/webx?14@@.3bbb579c/65 (http://www.adobeforums.com/cgi-bin/webx?14@@.3bbb579c/65)
One could define proper exposure as that exposure that optimizes capture information. For Ansel Adams and negative film, that involved exposing for the shadows. For digital capture, it involves exposing for the highlights. The nominal meter reading of the overall scene does not define proper expsure.
[a href=\"index.php?act=findpost&pid=65780\"][{POST_SNAPBACK}][/a]
I'm well aware of who Andrew is. If I'm not mistaken, he was asking a question and I was answering it. On that note, I'm not being dogmatic or overbearing. I'm simply stating what I know to work.
I have nothing to argue with Mr Fraser's quote -he's absolutely correct, as are you in your last paragraph there- but our discussion is not the one he was in. For the purpose of this discussion I am stating that over-exposure is where tones appear too bright, were 18% grey is lighter than 18% grey, were colors appear washed out, etc. Since this discussion has included post processing and not just capture, I'm referring to correct exposure as the final appearance of the image; what you'll take to print.
Just as you stated, "In ETTR, the image might be too light without exposure adjustment in the raw converter." I'm saying the same thing.
Thanks however, I'm still trying to ensure that all this expose to the right, use the white tile stuff is working and makes scenes to me and others. I'm in the same boat with the rest of you <g>.
If there were a way to photograph an object and objectively evaluate that I was placing the most bits in the last stop (and I'd have to recognize real data from noise), that would be useful.
[a href=\"index.php?act=findpost&pid=65785\"][{POST_SNAPBACK}][/a]
People are a good subject to test ETTR on. Have someone hold the white balance tool and make sure there some items or clothing that are dark toned and that there are some shadowed areas that will still relay some detail. Bracket photograph them and then see what appears the best.
I'm fairly confident that you'll see that a exposure in that test resulting in the whites being 250 the person will appear washed out and when corrected to 240, the person will appear as they should. You should also see a reduction of noise in the darker areas of the ETTR shots and sometimes -depending on the lens, camera resolution, photographic technique, etc- you might see some added sharpness in mid-tones to highlights.
-
I have nothing to argue with Mr Fraser's quote -he's absolutely correct, as are you in your last paragraph there- but our discussion is not the one he was in. For the purpose of this discussion I am stating that over-exposure is where tones appear too bright, were 18% grey is lighter than 18% grey, were colors appear washed out, etc. Since this discussion has included post processing and not just capture, I'm referring to correct exposure as the final appearance of the image; what you'll take to print.
[a href=\"index.php?act=findpost&pid=65802\"][{POST_SNAPBACK}][/a]
I don't see where you ever defined what you meant by proper exposure, but in Andrew's test he exposed according to the reading of his Minolta meter and brought the images into ACR without any exposure adjustment. Obviously, if you want 18% gray to appear as 18% gray in the print, you will have to adjust the exposure in ACR. However, in the context of ETTR, the image was not overexposed and Mr. Fraser was quoted in the proper context.
-
I don't see where you ever defined what you meant by proper exposure...
[a href=\"index.php?act=findpost&pid=65804\"][{POST_SNAPBACK}][/a]
My apologies if there was any confusion as to what I meant by "proper exposure" but now I have defined what I mean so people can understand what I was talking about.
If everyone here wants to define exposure how Fr. Fraser did then I'm down with that (as I said before, it's not wrong). However, if so, then how should setting exposure correction in the raw converter to set the proper luminosity of the final image be defined? How about "output luminosity?"
-
My apologies if there was any confusion as to what I meant by "proper exposure" but now I have defined what I mean so people can understand what I was talking about.
If everyone here wants to define exposure how Fr. Fraser did then I'm down with that (as I said before, it's not wrong). However, if so, then how should setting exposure correction in the raw converter to set the proper luminosity of the final image be defined? How about "output luminosity?"
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=65817\")
Output luminosity is fine with me. IMHO, exposure should be applied to what is done in the camera and is defined in lux seconds and can not be changed after the fact. We can argue what proper exposure is, but Bruce's definition of overexposure makes sense to me.
In terms of translating ISO settings to output luminosity, the ISO saturation standard for digital cameras indicates that an 18% gray card exposed according to the standard should read 18/106 of full scale, or a pixel level of 696 at the output of a 12-bit A-to-D converter, which can represent 4096 levels. Assuming a gamma of 2.2, the pixel value would be 114 with 8 bit output.
[a href=\"http://www.normankoren.com/digital_cameras.html]http://www.normankoren.com/digital_cameras.html[/url]
Julia Borg has a nice table relating reflection values and pixel values for various gammas:
http://www.pochtar.com/gamut_view/gamma.htm (http://www.pochtar.com/gamut_view/gamma.htm)
Of course, raw converters apply a tone curve as well as a gamma correction so that the output of ACR would be somewhat different.
-
A somewhat interesting update. I purchased a version of RAW Developer (Michael mentions it in his recent article "Measuring Megabytes"). I've played with a demo for a few months but decided that today with V1.5 out running native on Intel Mac's, I'd go for it. Nice product. Anyway, what's interesting is that bringing in the "correct" exposure Macbeth with BabelColor tile done yesterday, the default rendering produces nearly identical RGB values for white, gray and black in that RAW converter as ACR with Auto off. Other colors are different (the red is much better in RAW Developer*). I didn't expect the default renderings exposure wise to be so darn similar with different converters but when you think about it, that's what SHOULD be happening.
RAW Developer R152/G92/B62
ACR R182/G104/B72
-
Not to get too OT, but since were talking about the WhiBal and the BabelColor, what about the Expodisc gizmo?
-
Output luminosity is fine with me. IMHO, exposure should be applied to what is done in the camera and is defined in lux seconds and can not be changed after the fact.
This "exposure" you refer to is a very important one to consider. It is really the one that is the starting point for determining S/N ratios; once you have chosen this "absolute exposure" as I call it, then the choice of ISO is really counter-intuitive, based on popular belief. The highest gain-based ISO will give the least noise, unless the camera does a really horrible job of high-ISO amplification. The only issue with going high with the ISO is clipping (again, assuming absolute exposure is fixed). The choice of ISO with the manual exposure affects the "digitization depth" or the histogram.
In terms of translating ISO settings to output luminosity, the ISO saturation standard for digital cameras indicates that an 18% gray card exposed according to the standard should read 18/106 of full scale, or a pixel level of 696 at the output of a 12-bit A-to-D converter, which can represent 4096 levels.
[a href=\"index.php?act=findpost&pid=65824\"][{POST_SNAPBACK}][/a]
About 350 is what is used in practice (for green; for red much less). By the standard you mention, cameras are overstating their ISO by about 2x.
-
Bruce is your buddy and business partner--why don't you get his input and tell us why he prefers R4C2 in the CC?
[a href=\"index.php?act=findpost&pid=65704\"][{POST_SNAPBACK}][/a]
It's pretty simple really. The reason we white-balance rather than gray-balance is that the midpoint in a gamma 1.0 capture (level 128 in 8-bit terms) is a light gray around Lab L* 76. R2C4 on the Macbeth is a little lighter than this, R3C4 is quite a bit darker, so R2C4 is the closest patch to the midpoint.
-
This "exposure" you refer to is a very important one to consider. It is really the one that is the starting point for determining S/N ratios; once you have chosen this "absolute exposure" as I call it, then the choice of ISO is really counter-intuitive, based on popular belief. The highest gain-based ISO will give the least noise, unless the camera does a really horrible job of high-ISO amplification. The only issue with going high with the ISO is clipping (again, assuming absolute exposure is fixed). The choice of ISO with the manual exposure affects the "digitization depth" or the histogram.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=67087\")
John,
Your post is a bit confusing. By gain I gather you are referring to amplifier gain, which is increased in high ISO settings. CCD system gain is reported in terms of electrons per ADU (alalog to digital unit) and is the inverse of amplifier gain.
While a CCD gain of 1 electron / ADU would be perferable, this would require 16 bit encoding with most current 35mm style cameras, which can record 50,000 electrons easily. The full well of the Canon EOS 1D is ~79,000 electrons and a gain of 1 e/ADU would require 17 bits (inconvient).
While total noise in absolute terms increases with increasing exposure up to full well (according to Poisson sampling), the signal to noise ratio is best at full well. See the discussion by Roger Clark.
[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]
For optimum S/N one should expose so that the sensor is full well in the highlights and this is related to lux/seconds and not the ISO chosen. If you expose for full well and use an ISO higher than base, clipping will occur with 12 bit AD converters unless the camera has a setting to vary CCD gain, which is usually available only on specialized cameras for scientific purposes. I think that ISO 50 on some Canons sets the gain to 1/2x, and there is a risk of blowing highlights with this setting.
http://www.photomet.com/library_enc_gain.shtml (http://www.photomet.com/library_enc_gain.shtml)
If you expose at less than full well, then the amplifier gain can be increased. Absolute noise will be reduced, but so will S/N, and this is not what most of us want.
-
John,
Your post is a bit confusing.
I don't think so. My post makes a very reasonable assumption that you are in the real world, and can't come close to saturating the sensor in many situations, without getting motion blur, or shallower DOF than desired. That is where choosing aperture and shutter speed come in.
By gain I gather you are referring to amplifier gain, which is increased in high ISO settings. CCD system gain is reported in terms of electrons per ADU (alalog to digital unit) and is the inverse of amplifier gain.
While a CCD gain of 1 electron / ADU would be perferable, this would require 16 bit encoding with most current 35mm style cameras, which can record 50,000 electrons easily. The full well of the Canon EOS 1D is ~79,000 electrons and a gain of 1 e/ADU would require 17 bits (inconvient).
Again, this is the real world. The current DSLRs can barely deliver 9 or 10 bits worth of really useful information. A nice, clean 16- or 17-bit readout is a pipedream at this point in time for most people. Maybe some of the expensive MF backs do better, but base ISO perfomance on most cameras is abysmal.
Also, I think that if in some miracle, a way of reading/digitizing sensors that only includes poisson and dark current noise were made available, that there would be any reason to stop at approximately 1 ADU = 1 electron. Another bit or two would keep the posterization noise down. I wish we had that problem. A clean, 16 or 18-bit digitization of the sensor charges would eliminate the need to change ISOs; ISO 100 under-exposed by 4 stops would be as good or better than ISO 1600 is now, even with the same well capacity.
While total noise in absolute terms increases with increasing exposure up to full well (according to Poisson sampling), the signal to noise ratio is best at full well. See the discussion by Roger Clark.
I've done this research myself. I have a spreadsheet with the noise values at various RAW levels at various ISOs from my 20D; I used this data to determine if other people with 20Ds and 30Ds, especially, really had noise problems in their cameras. In every case it turned out that the noise levels were the same in these cameras, and that it was either exposure or conversion that was emphasizing noise.
For optimum S/N one should expose so that the sensor is full well in the highlights and this is related to lux/seconds and not the ISO chosen. If you expose for full well and use an ISO higher than base, clipping will occur with 12 bit AD converters unless the camera has a setting to vary CCD gain, which is usually available only on specialized cameras for scientific purposes.
Again, you are assuming that shutter speed and aperture are not real issues. They are really the most important thing to choose in many situations. Full exposure at ISO 100 is not an option. For a lot of wildlife or spontaneous night-shooting in city streets, a full histogram at ISO 1600 isn't even possible.
I think that ISO 50 on some Canons sets the gain to 1/2x, and there is a risk of blowing highlights with this setting.
ISO 50 is probably implemented as an arithmetic trick in most cases, which loses a stop of highlights, relative to the metering. In some, ISO 100 doesn't use full-well (Canon 1dmkII), and 50 actually uses a little more of the sensor's DR.
If you expose at less than full well, then the amplifier gain can be increased. Absolute noise will be reduced, but so will S/N, and this is not what most of us want.
[a href=\"index.php?act=findpost&pid=67092\"][{POST_SNAPBACK}][/a]
No, that's not what we usually want, but that is often unavoidable.
Anyway, you seemed to miss my point - Once you have decided what aperture and shutter speed you want to use (low sensor signal is the primary cause of noise), the *highest* ISO will give the least noise, especially in the shadows (clipping being the only danger).
Let's say you take a shot at ISO 400, and you review the histogram, and notice that there is a stop of unused highlights in the histogram. You can fill that histogram in two ways; by decreasing the f-stop and/or increasing the exposure time, or by doubling the ISO (some cameras have intermediate f-stops, but they are not always real). The former will decrease noise, but you might be sacrificing sharpness or DOF, or the lens' sweet spot. Increasing the ISO will also reduce the noise, not quite as much as increasing the sensor exposure (but more than you might think), but there is no compromise on the Tv and Av settings you want to use.
You seem to be replying to me as if I had written "You get the lowest noise by using the full RAW DR, regardless of ISO", which I did not write. My original statement was one with a condition - that you have aready chosen the f-stop and shutter speed you want. There is no option for full-well exposure in many cases of the condition.
-
Once you have decided what aperture and shutter speed you want to use (low sensor signal is the primary cause of noise), the *highest* ISO will give the least noise, especially in the shadows (clipping being the only danger).
[a href=\"index.php?act=findpost&pid=67118\"][{POST_SNAPBACK}][/a]
This is a very significant point. I've got lots of ruined shots as a result of giving undue weight to 'absolute' noise in an image. Unfortunately, it's been a slow learning process. I should have used ISO 400 and even 800 more often with my D60. I'm not going to make the same mistake with my 5D. If the shot requires ISO 1600 for an adequate shutter speed and DoF, I'll use it.
-
This is a very significant point. I've got lots of ruined shots as a result of giving undue weight to 'absolute' noise in an image. Unfortunately, it's been a slow learning process. I should have used ISO 400 and even 800 more often with my D60. I'm not going to make the same mistake with my 5D. If the shot requires ISO 1600 for an adequate shutter speed and DoF, I'll use it.
[a href=\"index.php?act=findpost&pid=67150\"][{POST_SNAPBACK}][/a]
With the Canon 20D, 1 stop of under-exposure ("under" means "not to the right", not just "dark") results in more shadow noise than ISO 800 fully exposed. The 1DmkII is similar, and your 5D probably is, too.
It is rather unfortunate that the majority of digital users believe that "higher ISOs add more noise", and that only a very dark preview image at a low ISO would benefit from a higher ISO with the same absolute exposure. An image that looks fine in the LCD could easily have 2 stops of unused RAW highlights.
What I long for in future cameras is manual exposure (Av and Tv) with floating ISO, with either a "highlight escape control" that is exponential, like .01%, .1%, 1%, 10%, or a simple EC control as is currently used in auto-exposure modes. Such a paradigm would be truer to the way a photographer should really make decisions, and the way digital exposure really works. One of the biggest complaints about such a possible mode is that there is a loss of control of image quality with auto-ISO. Such complaints are based on a misconception about the role of ISO in noise - ISO only increases noise when it reduces exposure through metering. Manual Tv and Av with auto-ISO does not have that problem; having the ISO in the viewfinder, optionally flashing at the higher ISOs, would tell the photographer that it is possible to sacrifice f-stop or shutter speed for a cleaner image at a lower ISO.
-
With the Canon 20D, 1 stop of under-exposure ("under" means "not to the right", not just "dark") results in more shadow noise than ISO 800 fully exposed. The 1DmkII is similar, and your 5D probably is, too.
It is rather unfortunate that the majority of digital users believe that "higher ISOs add more noise", and that only a very dark preview image at a low ISO would benefit from a higher ISO with the same absolute exposure. An image that looks fine in the LCD could easily have 2 stops of unused RAW highlights.
What I long for in future cameras is manual exposure (Av and Tv) with floating ISO, with either a "highlight escape control" that is exponential, like .01%, .1%, 1%, 10%, or a simple EC control as is currently used in auto-exposure modes. Such a paradigm would be truer to the way a photographer should really make decisions, and the way digital exposure really works. One of the biggest complaints about such a possible mode is that there is a loss of control of image quality with auto-ISO. Such complaints are based on a misconception about the role of ISO in noise - ISO only increases noise when it reduces exposure through metering. Manual Tv and Av with auto-ISO does not have that problem; having the ISO in the viewfinder, optionally flashing at the higher ISOs, would tell the photographer that it is possible to sacrifice f-stop or shutter speed for a cleaner image at a lower ISO.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=67161\")
The whole noise issue can be summed up by examining the analysis posted by Roger Clark on his website. His data are for the Canon EOS 1D Mark II.
[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]
N = (P + r2 + t2)1/2, (eqn 2)
Where N = total noise in electrons, P = number of photons, r = read noise in electrons, and t = thermal noise in electrons
[/b]
Since photon noise is dominant under normal shooting conditions, one should expose so that a maximum number of photons fall on the sensor. If considerations of f/stop and shutter speed preclude exposing to full well, one should give as much exposure as conditions permit. One then has to choose an ISO that will give optimum results.
Current amplifiers introduce very little noise, so from this standpoint, it does not make much difference whether one chooses ISO 100 and boosts the signal in PP or ISO 1600 for exposure to the right, which will, however, make maximum use of the full range of the analog to digital converter and give smaller quantization errors.
Read noise must then be considered. While photon noise is predominant in the highlights, read noise is often predominant in the shadows. One would then choose an ISO that gives the lowest read noise. From Dr. Clark's analysis, ISO 100 has a read noise of 16.6 electrons and ISO 1600 has a read noise of 3.9 electrons. Obviously, the ISO 1600 will not only result in less quantization error but also less read noise, so it is the obvious choice. However, at ISO 3200 read noise increases and dynamic range is reduced, so it is best to stick with ISO 1600. ISO 800's read noise is similar to that of ISO 1600 and ISO 800 is not a bad choice either.
Thermal noise becomes prominent only with very long exposures
-
What I long for in future cameras is manual exposure (Av and Tv) with floating ISO, with either a "highlight escape control" that is exponential, like .01%, .1%, 1%, 10%, or a simple EC control as is currently used in auto-exposure modes. Such a paradigm would be truer to the way a photographer should really make decisions, and the way digital exposure really works. One of the biggest complaints about such a possible mode is that there is a loss of control of image quality with auto-ISO. Such complaints are based on a misconception about the role of ISO in noise - ISO only increases noise when it reduces exposure through metering. Manual Tv and Av with auto-ISO does not have that problem; having the ISO in the viewfinder, optionally flashing at the higher ISOs, would tell the photographer that it is possible to sacrifice f-stop or shutter speed for a cleaner image at a lower ISO.
[a href=\"index.php?act=findpost&pid=67161\"][{POST_SNAPBACK}][/a]
Great idea, John. I second the motion.
But once Canon engineers latch onto the idea, they'll probably want to bury it eleven levels deep in the menu structure, so that it is harder to use than their infamous "mirror lockup" feature.
Eric
-
N = (P + r2 + t2)1/2, (eqn 2)
Sure that wasn't (P+r2+t2)^(1/2)? Noise doesn't add linearly.
Obviously, the ISO 1600 will not only result in less quantization error but also less read noise, so it is the obvious choice. However, at ISO 3200 read noise increases and dynamic range is reduced, so it is best to stick with ISO 1600. ISO 800's read noise is similar to that of ISO 1600 and ISO 800 is not a bad choice either.
[a href=\"index.php?act=findpost&pid=67168\"][{POST_SNAPBACK}][/a]
3200 is non-existent on most cameras, except as an arithmetic trick. Noise should be double what it is relative to metering, and the same in electrons, as ISO 1600.
-
Indeed, and it is not just a matter of noise reduction: underexposing by four stops at ISO 100 means that the signal passes through the last pre-amplifier stages 1/16th as strong, and then has to be amplified 16 times as much in the digital domain to get correct levels. This means that an noise introduced late on the pre-amplifier stage and "quantization noise" from A/D conversion gets amplified by an extra factor of 16.[a href=\"index.php?act=findpost&pid=62976\"][{POST_SNAPBACK}][/a]
I believe that Ray may have been talking about the phenomenon I demonstrated in the pre-pay Galbraith forums. I took a shot with a wide-angle lens from a dock at dusk of pilings, in the sound, under-exposed for ISO 1600, in manual mode, and then took the same shot dialing the ISO down to 100 without changing the aperture and shutter speed. I showed the RAW data from both, at the same scale, interpolated, and with WB. One was 1600 pushed to 10,000, and the other was 100 pushed to 10,000. The 1600 setting provided the clearest image, by far. At one time, I would assume that this was due to posterization (1/16 as many levels for the ISO 100), but I decided to posterize the 1600 image so that it had the same number of levels to represent it as the 100 image (20 out of 4096 levels, from 320). To my surprise, the 1600 image lost almost no image quality whatsoever, when posterized. It was still orders of magnitude cleaner than the 100 image. Conclusion - under-exposure at ISO 100 resulted in sloppy images not because of posterization, but because of tremendous readout noise.
-
To my surprise, the 1600 image lost almost no image quality whatsoever, when posterized. It was still orders of magnitude cleaner than the 100 image. Conclusion - under-exposure at ISO 100 resulted in sloppy images not because of posterization, but because of tremendous readout noise.
[a href=\"index.php?act=findpost&pid=67232\"][{POST_SNAPBACK}][/a]
John,
We should also bear in mind that noise at base ISO in Canon DSLRs remains fairly constant from model to model, which is reflected in dpreview's noise test charts. If the 1Ds produces less noise (in the final image) at ISO 100 than the older D60, it's only because the 1Ds has a greater number of pixels. On a pixel for pixel basis, D60 noise is actually marginally less.
Canon appears to think that noise at ISO 100 is already good enough (and provided the DR of the scene is not excessive, they are probably right). The early models of Canon DSLRs, including the D30 and 1D, did not show this 'lower read-out noise' at higher ISOs. The 10D set the trend to reduced noise at higher ISOs and it's just got better with successive models.
Are you sure that reduced read-out noise is the only factor here? If so, what are the obstacles to reducing read noise at ISO 100?