Camera-to-Print -- exposing to the right question

[Panopeeper]

Two notes to the above.

1. Higher ISO does not mean automatically more noise.  saw a demonstration on DPReview (one of the authors was John Sheehy) proving, that not the higher ISO but low light causes noise.

The consequence of this is, that in such situations, when the lighting is enough but the exposure can not be increased to achieve ETTR because of other considerations, then increasing the ISO helps gaining more detail in the shadows.

Higher ISO is certainly no substitute for higher exposure - when the latter is possible.

2. The optimal ISO is not always the base. Example: the Canon D40. The DR with ISO 200 is about 0.2 EV larger than that of ISO 100. This is due to the fact, that the range of pixel values is larger with ISO 200 than with ISO 100. 0.2 EV is not much wort (though it could be important), but perhaps other cameras will follow suit and offer a much higher value range with higher ISOs. 14 bits should be plenty enough for that.

[bjanes]

Two notes to the above.

1. Higher ISO does not mean automatically more noise.  saw a demonstration on DPReview (one of the authors was John Sheehy) proving, that not the higher ISO but low light causes noise.

The consequence of this is, that in such situations, when the lighting is enough but the exposure can not be increased to achieve ETTR because of other considerations, then increasing the ISO helps gaining more detail in the shadows.
[a href=\"index.php?act=findpost&pid=165140\"][{POST_SNAPBACK}][/a]

I don't know what the above is in reply to, but my analysis clearly showed that noise is related almost entirely to exposure, not ISO. However, you can not expose at full well much above base ISO because you will get clipping in the ADC. Some cameras place the highlights some distance below ADC clipping, and you might be able to increase ISO somewhat and still expose at the full well of the sensor if you have one of those. In addition, there is often less read noise at high ISO and that could improve the S:N in the deep shadows. However, highlight S:N would be worse owing to the effect of shot noise.

[larsrc]

2. The optimal ISO is not always the base. Example: the Canon D40. The DR with ISO 200 is about 0.2 EV larger than that of ISO 100. This is due to the fact, that the range of pixel values is larger with ISO 200 than with ISO 100. 0.2 EV is not much wort (though it could be important), but perhaps other cameras will follow suit and offer a much higher value range with higher ISOs. 14 bits should be plenty enough for that.
[a href=\"index.php?act=findpost&pid=165140\"][{POST_SNAPBACK}][/a]

I get the impression that optimal ISO does not vary from one instance of a particular model to another.  Is there a database of optimal ISOs for various (SLR) cameras somewhere? Could we make one?  I could certainly do the database setup.

-Lars

[John Sheehy]

...you'll get a better result by increasing the exposure a little bit to the right, which is the same thing as lowering the ISO...
you shouldn't follow the 'expose to the right' religion if you're shooting at high ISO. there's no point in shooting at ISO 400 and exposing to the right. you may as well drop to ISO 200.

I don't understand this. the purpose of exposing to the right is to get more levels in the middle of the histogram because the first stop provides 4096 levels, the second stop 2048, etc. if you shoot at ISO 800 and expose to the right, you're getting more levels for the middle and low regions. how does this equate to shooting at ISO 400 without exposing to the right in terms of levels and detail? i.e., ISO 800 ev +1.0 == ISO 400 ev 0.0

The number of levels in a stop is not necessarily the real issue.  It works as a model because it parallels the real issue; signal-to-noise ratios.  Also, in theory, if there were no noise or very little noise, then more levels would avoid RAW quantization, but that doesn't seem to be an issue with the RAW data in current cameras; all have far more than enough levels for the highlights and midtones, and 12 bits is just good enough for the best ISO 100s out there, in the deepest shadows (the first place RAW quantization would show).  When you start comparing different relative (but similar absolute) exposures at different ISOs, things get a little bit more complex.  If you expose manually, and set the Av and Tv values, and vary the ISO, the optimal ISO may be a low one, or the highest one that doesn't clip desired highlights, depending on the camera.  Some cameras have horrible amplifiers in them for high ISO, and ruin the high ISOs.  They could give better results by shooting at a lower ISO and under-exposing.  This is certainly not true of your Canon DSLR, though.  Canon DSLRs, except for the original 1D, add less noise to the signal (relative to it) at the highest ISOs.

please explain. details and levels in the lower region of the histogram are important to me because many of the birds I photograph have dark regions which don't contain as much detail as I would prefer.

and... is this level structure carried into the RAW processor and beyond? i.e., after processing, is the lowest stop still restricted to only 128 levels?

"The lowest stop" most likely refers to the lowest stop under consideration or considered to be useful, and everything below it, as a unit.  There really is no "lowest stop", per se, except for a few cameras that clip slightly above the true black level in their RAW data.  The only issue as you go down through all these stops is that the noise gets stronger and stronger, relative to the weak signal There is no great difference in kind between a signal being slightly below or slightly above the noise floor; the SNR is just different by degree.  There is signal below the noise floor.

and... the Canon Mark III shoots in 14-bit (from Canon's page: "The extra power of dual DIGIC III processors has also allowed analog-to-digital conversion to improve from 12 to 14 bits per channel, meaning better tonal gradation for RAW images."). would this increase the levels of the lower stop in the histogram by 2^2 from 2^7 to 2^9 == 512 levels?
[a href=\"index.php?act=findpost&pid=132070\"][{POST_SNAPBACK}][/a]

The amount of IQ benefit that comes from 14-bit data is maximum at the lowest ISOs, and is nearly infinitesimal.  There is too much analog noise before and during digitization for the extra bits to carry any significant extra signal.  However, being in a format with a higher precision, the extra bits may force converters to use more precision in the conversion process; something that could be done, also with 12-bit data.

[Mark D Segal]

The amount of IQ benefit that comes from 14-bit data is maximum at the lowest ISOs, and is nearly infinitesimal.  There is too much analog noise before and during digitization for the extra bits to carry any significant extra signal.  However, being in a format with a higher precision, the extra bits may force converters to use more precision in the conversion process; something that could be done, also with 12-bit data.
[{POST_SNAPBACK}][/a]

You may find this research of interest: [a href=\"http://tinyurl.com/2e4nss]12 vs 14 bit[/url]

Now, let us set aside the question of IQ differences between 12 and 14 bit (though Canon claims more levels = somoother tonal gradations - in principle should be true, but the in practice the issue is whether or not it is or one day will be noticeable).

I would like to gain clarity on clarity    if I may. Let us accept that ETTR (without clipping desired highlight detail) is "a good thing" because the extra exposure improves S/N visibly especially in the three-quarter tones and below and should provide smoother tonal gradations because it puts those tones further to the right in the histogram where the number of levels is higher (i.e. more exposure). Then we rebalance the image to taste in the raw converter - Lightroom, ACR or whatever.

The question on which I would appreciate more clarity is how best to do ETTR when the camera's automatic metering is not doing so. As I see it, I have two choices (switching to manual exposure); (i) increase the ISO leaving aperture and shutter speed unchanged, or (ii) increase exposure by leaving the ISO unchanged and either opening the aperture or slowing the shutter speed. Let us say compensation worth 1 to 1.5 stops is generally needed to achieve this (though it could vary by scene). Which of these approaches (i.e. (i) or (ii) ) will give me an image with less visible noise and why?

[Ray]

The question on which I would appreciate more clarity is how best to do ETTR when the camera's automatic metering is not doing so. As I see it, I have two choices (switching to manual exposure); (i) increase the ISO leaving aperture and shutter speed unchanged, or (ii) increase exposure by leaving the ISO unchanged and either opening the aperture or slowing the shutter speed. Let us say compensation worth 1 to 1.5 stops is generally needed to achieve this (though it could vary by scene). Which of these approaches (i.e. (i) or (ii) ) will give me an image with less visible noise and why?
[a href=\"index.php?act=findpost&pid=167463\"][{POST_SNAPBACK}][/a]

Mark,
The way I see it is, there's a primary requirement for a specific aperture for DoF purposes and a primary requirement for a shutter speed sufficient to freeze subject movement and/or camera shake.

In order to obtain those two requirements, an appropriate ISO should be selected.

Whatever ISO is selected, it is usually advantageous to expose to the right, especially with Canon DSLRs, but not apparently with some models of cameras.

I recall Edmund writing in another thread that he thought he was getting better results underexposing a P45+ DB by 3 stops at ISO 100, rather than using the same exposure at ISO 800 which would result in an ETTR in relation to that ISO setting.

My own experiments with the 20D and 5D have demonstrated very clearly that an ETTR at a higher ISO is always better than the same exposure at a lower ISO setting. However, if you can be flexible about the choice of either aperture or shutter speed, the lower ISO will generally produce better results, provided it's an ETTR.

In order to achieve accurate ETTR it may be necessary to bracket exposure or to adopt a more time-consuming method in manual mode. One such method, if your camera has a spot meter mode, is to take a reading of the brightest part of the scene, then increase exposure by 3 stops.

Another method is to take the shot, examine the histogram and if the assessment is that it is not far enough to the right, take the shot again at a higher ISO such that the histogram is pushed more to the right.

In order to get the appearance of the histogram and the highlight warning on the preview image as accurate as possible, it's probably advisable to reduce contrast, saturation and sharpening to a minimum in the camera's menu. However, this is a matter for experimentation. My own cameras are set up so that a very small amount of highlight flashing is just right for a full ETTR.

[John Sheehy]

You may find this research of interest: 12 vs 14 bit

The 12-bit and 14-bit readouts of the D300 might as well be on different cameras.

6fps vs 2.5 tells you that there is a lot more going on than bit depth here.  12-bit is most likely a quick'n'dirty readout, quite a different than than if the 2.5 fps readout only recorded 12 bits in the RAW.

The question on which I would appreciate more clarity is how best to do ETTR when the camera's automatic metering is not doing so. As I see it, I have two choices (switching to manual exposure); (i) increase the ISO leaving aperture and shutter speed unchanged, or (ii) increase exposure by leaving the ISO unchanged and either opening the aperture or slowing the shutter speed. Let us say compensation worth 1 to 1.5 stops is generally needed to achieve this (though it could vary by scene). Which of these approaches (i.e. (i) or (ii) ) will give me an image with less visible noise and why?
[a href=\"index.php?act=findpost&pid=167463\"][{POST_SNAPBACK}][/a]

My paradigm (even if not directly supported by the camera) is to choose real world exposure based on a compromise of photon collection, shutter speed, and aperture, and choose the highest ISO that won't clip desired highlights, because the highest ISO will give the least "added" read noise to the absolute exposure (on most cameras, to a small or great degree).  The real tradeoff isn't between ISO and anything; the real tradeoff is absolute exposure.  IQ only parallels ISO when you are exposing the same relative way in each ISO.

[Mark D Segal]

Interesting, both you and Ray are saying that higher ISOs are less noisy - specifically in John's post - less read noise. But do they not also produce more "shot" noise? Is it not normally the case that we see more noise in images shot at high ISO?

Ray - some qualifications on aspects of your suggestions, I think. Firstly, if the camera is in an auto exposure mode, changing ISO inversely changes the exposure so one shouldn't get closer or further to/from the desired ETTR setting unless one bumps into a constraint. If the approach is to change ISO, I do believe one must move into manual mode to preserve the exposure settings and move the histogram rightward, or again in manual mode over-ride the auto setting at a given ISO and expose more.  (John, I believe, perhaps incorrectly?, that there are trade-offs between these options.)

Secondly, when shooting raw of course, contrast, saturation and sharpening is not relevant - it should all be turned off/neutral so one imports an image as raw as it gets from the camera to the raw converter. Working from camera-baked jpegs I can see may be another story.

[bernie west]

Just when I thought I had ettr worked out, I came across this thread, and it seems there is more to it.  It seems as if the original concept (which I have been following) as proposed by Reichmann has progressed some.  Can the argument be summarized like this:

Shooting above base iso and ettr may not give you are better SNR than shooting the same exposure at base iso (even though the histogram may only reach half way along the scale)?

If that's the case then clearly we are better shooting at base iso, than employing analogue gain via iso boost.  But obviously at some point (say iso 800 vs iso 100) the iso 100 shot is going to be underexposed and will require boosting via the raw converter.  At this point then it must become a case of diminishing returns, as boosting continuous analogue signal should always be better than boosting discrete digital data.  Am I right in this observation?

I did a test at iso 400 vs 100.  Iso 400 was ettr, and pulled back -1ev in DPP, and iso 100 was pushed 1ev.  Both were shot with the same absolute exposure (ie. same Av and Tv).  The following images show a 100% crop from both (iso 400 on the left).  You can clearly see more noise in the iso 100 +1ev vs the iso 400 -1ev.

[Ray]

Ray - some qualifications on aspects of your suggestions, I think. Firstly, if the camera is in an auto exposure mode, changing ISO inversely changes the exposure so one shouldn't get closer or further to/from the desired ETTR setting unless one bumps into a constraint. If the approach is to change ISO, I do believe one must move into manual mode to preserve the exposure settings and move the histogram rightward, or again in manual mode over-ride the auto setting at a given ISO and expose more.

Mark,
Yes, of course you're right. I forgot to mention that. It's necessary to go into manual mode and here's a situation which can be a bit of a hassle. Getting into manual mode is easy with a turn of a dial. Changing ISO is easy with a press of a button and a turn of a dial. But one also has to change the f stop and shutter speed in manual mode. One could easily miss the shot with all that stuffing around.

Michael has made a big deal of the lack of a dedicated MLU button on Canon cameras. How about a button that gets you into manual mode whilst preserving the aperture and shutter speed that was last used in the previous mode. Perhaps there's a way of programming some cameras to do this, perhaps even my own camera. I sometimes miss the obvious.

Secondly, when shooting raw of course, contrast, saturation and sharpening is not relevant - it should all be turned off/neutral so one imports an image as raw as it gets from the camera to the raw converter. Working from camera-baked jpegs I can see may be another story.

It's relevant to the appearance of the LCD preview and the histogram since that is based on an in-camera jpeg conversion (so I'm told). My experience is, the amount of highlight warning flashing that takes place is significantly affected primarily by the contrast setting. All you have to do is try shooting an ETTR of the same scene with contrast at a minimum and contrast at a maximum. When contrast is at a maximum, you could have an entire sky flashing in the preview. But the same exposure, with in-camera contrast set at a minimum, will produce a preview with just a small area of the brightest part of the sky flashing.

[Mark D Segal]

My experience is, the amount of highlight warning flashing that takes place is significantly affected primarily by the contrast setting. All you have to do is try shooting an ETTR of the same scene with contrast at a minimum and contrast at a maximum. When contrast is at a maximum, you could have an entire sky flashing in the preview. But the same exposure, with in-camera contrast set at a minimum, will produce a preview with just a small area of the brightest part of the sky flashing.
[a href=\"index.php?act=findpost&pid=168481\"][{POST_SNAPBACK}][/a]

Ray, this is interesting - do you find in general the minimum contrast setting produces an impression of clipping that is closer to what you would eventually observe in the raw converter?

[John Sheehy]

Interesting, both you and Ray are saying that higher ISOs are less noisy - specifically in John's post - less read noise. But do they not also produce more "shot" noise? Is it not normally the case that we see more noise in images shot at high ISO?

What I meant is that they are less noisy in an absolute sense.

IOW, for a given real world exposure, higher ISOs add less noise to the RAW data.  If you are using a fixed set of Tv and Av values, and just varying the ISO, shot noise does not vary.  Shot noise has nothing directly to do with the ISO setting; it is only related in the sense that a high ISO clips away the sensor ranges of exposure where the lowest shot noise occurs.

In simple theory, shot noise should be the only noise of significance in the midtones and highlights, but read noise is usually kind of nasty, in the sense that it isn't totally random, and my have some patterns which make it much more visible than its statistical strength might suggest.  Case in point, the other guy's test this evening.  He is looking at midtones, but with the same absolute exposure, with the same shot noise, he got significantly more visible noise in the low ISO shot.

Ray - some qualifications on aspects of your suggestions, I think. Firstly, if the camera is in an auto exposure mode, changing ISO inversely changes the exposure so one shouldn't get closer or further to/from the desired ETTR setting unless one bumps into a constraint. If the approach is to change ISO, I do believe one must move into manual mode to preserve the exposure settings and move the histogram rightward, or again in manual mode over-ride the auto setting at a given ISO and expose more.

Well, you can do that just as quickly sometimes with the EC control.

(John, I believe, perhaps incorrectly?, that there are trade-offs between these options.)

I'm not sure I'm parsing your grammar there, but the general idea is that whatever amount of light you decide to capture, the best ISO is the one that comes just short of clipping desired highlights.  This is very valuable with Canon DSLRs, the more recent Nikons, and possibly others I'm not aware of, and it applies to a smaller degree to older Nikons and most other non-Canon DSLRs and P&S cameras.  There are indeed some cameras out there that do worse by using high ISOs than by under-exposing at low ISOs.  My Panasonic FZ50 is like that.  For any ISO greater than 200, it is better to shoot RAW and under-expose ISO 200.  It seems to use a really nasty amplifier to achieve high ISOs.

Secondly, when shooting raw of course, contrast, saturation and sharpening is not relevant - it should all be turned off/neutral so one imports an image as raw as it gets from the camera to the raw converter. Working from camera-baked jpegs I can see may be another story.
[a href=\"index.php?act=findpost&pid=167560\"][{POST_SNAPBACK}][/a]

Well, you can apply the same principles to JPEG, but JPEG will just have more limitations on how far you can take things.  I often use the 2.5 MB embedded JPEGs to make small images, even though I shot in RAW, and exposed to the right, but I have the JPEG contrast set to minimum, and the saturation at -1, and I can pull down ETTR JPEGs by lowering the gamma to 0.8 in post-processing.

[Panopeeper]

I often use the 2.5 MB embedded JPEGs to make small images

You have a 30D, don't you? The embedded image is 1728x1152, which is ok for that purpose, but all sizes I have seen were in the range 260KB-500KB, i.e. the quality can't be that great.

[Ray]

Ray, this is interesting - do you find in general the minimum contrast setting produces an impression of clipping that is closer to what you would eventually observe in the raw converter?
[a href=\"index.php?act=findpost&pid=168492\"][{POST_SNAPBACK}][/a]

Mark,
Yes. I find the Landscape Picture Style on my 5D set to minimum contrast results in a more reliable preview and histogram for assessing ETTR. Small areas of highlight flashing, sometimes barely discernible, are usually about right for ETTR. That's my guide. No flashing at all usually means underexposure. Significant flashing means overexposure.

[francois]

Ray, this is interesting - do you find in general the minimum contrast setting produces an impression of clipping that is closer to what you would eventually observe in the raw converter?
[{POST_SNAPBACK}][/a]

Andrew Rodney, in his [a href=\"http://www.digitalphotopro.com/index2.php?option=com_content&task=view&id=316&pop=1&page=0&Itemid=41]Exposing for RAW[/url] article, has someting interesting to say about contrast & settings:

…
Unfortunately, at least with my Canon EOS 5D, the “best” exposure often produced
 the worst-appearing LCD previews.
…
When I set my 5D Picture Style from the default to -4 contrast, the clipping of highlights
 on the LCD didn’t appear until I had “overexposed” 2⁄3 of a stop. This is better than
the original default settings, yet in actuality, I was able to overexpose 1! stops beyond
what my meter suggested while fully retaining  highlight data in my RAW file. The clipping
indicators are still far from correctly describing what’s  happening to the RAW data.
…

[Mark D Segal]

Ray and Francois,

Thanks - this means one thing and one thing only: experiment with a range of settings, make notes at the time the image is processing, then put them up in Camera Raw and see what works best. It could be that the settings which work well enough for most situations with one camera model are not satisfactory for another.