Pages: [1] 2 3   Go Down

Author Topic: Noise About Noise  (Read 23591 times)

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Noise About Noise
« on: June 03, 2008, 09:02:03 pm »

Mark Segal's post on noise is a valuable contribution for those interested in noise in digital images since it contains actual photographic comparisons and complements the more theoretical post by Emil Martinec.

One point that I would like to make is that ACR's histograms are a poor indication of the raw data, since ACR uses a baseline exposure offset of +0.5 EV for the D3 and with default settings the highlights will appear clipped when they are intact in the actual raw file. I would recommend Rawanalyze to look at the actual raw histogram.

Here is the raw histogram of a Stouffer stepwedge exposed so the highlights are just shot of clipping as shown with Rawnanalyze:



And here is what it looks like in ACR with default settings. The top two steps (each is 0.1 EV) appear clipped:



One needs to use -0.6 EV of negative exposure compensation to place the highlights just short of clipping:



I would like to ask Mark if his D3 shots are really properly exposed according to ETTR.

Bill
« Last Edit: June 03, 2008, 09:25:35 pm by bjanes »
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #1 on: June 03, 2008, 09:47:19 pm »

Quote
Mark Segal's post on noise is a valuable contribution for those interested in noise in digital images since it contains actual photographic comparisons and complements the more theoretical post by Emil Martinec.

One point that I would like to make is that ACR's histograms are a poor indication of the raw data, since ACR uses a baseline exposure offset of +0.5 EV for the D3 and with default settings the highlights will appear clipped when they are intact in the actual raw file. I would recommend Rawanalyze to look at the actual raw histogram.

Here is the raw histogram of a Stouffer stepwedge exposed so the highlights are just shot of clipping as shown with Rawnanalyze:

And here is what it looks like in ACR with default settings. The top two steps (each is 0.1 EV) appear clipped:

One needs to use -0.6 EV of negative exposure compensation to place the highlights just short of clipping:

I would like to ask Mark if his D3 shots are really properly exposed according to ETTR.

Bill
[a href=\"index.php?act=findpost&pid=199636\"][{POST_SNAPBACK}][/a]

Hi Bill,

Many thanks for reading and commenting. Appreciated. Some points:

First - something went wrong with the links to your illustrations - the first two started as the same image in Camera Raw 4.4., then as I was writing, the first two sorted out correctly. Weird.

Second, in the 3rd illustration is there any reason why you would prefer to reduce exposure rather than increase recovery to rescue the blown highlights?

Third, I like your idea of seeing these histograms in more than one raw converter. That could help shed light on the role of the converter in exposure rendition.

Fourth, the kind of histogram you get in the raw converter of course depends on what the user decides to be the ACR defaults. I notice you have Brightness and Contrast set on the usual ACR defaults we download the program with. Perhaps in the defaults you are using there is also a moderate S in the Point Curve. These things can cause some highlight clipping when the highlights are very close to the edge in the raw file. I turn all that stuff off to zero in the basic tab and set the tone curves to Linear. Then I see things as they are without any "pumping-up" in the raw converter. It's what I like to think of as a "Greenfield" starting point. Then I build up. So with that background I can turn to your question: were the D3 shots properly exposed as ETTR?

The answer is a qualified YES, in the sense that when we made the D3 shots, on the camera's LCD screen the histograms looked bang-on ETTR, except for minor clipping of specular highlights. Downloaded into ACR with the settings at the flat defaults I discussed just above (and in footnote 6 in the article), lo and behold there was clipping of the brightest portion of the sky, which we did NOT see in the camera LCD. However, using ACR's Recovery function, the blueness of the sky got perfectly recovered. So I would say the D3 shots were ETTR+a bit.

I think perhaps one lesson of experience from doing this work is the usefulness of experimenting to select in-camera JPEG settings and raw converter default settings that will best produce similar histograms of the raw file in both places - should limit unpleasant surprises.
« Last Edit: June 03, 2008, 09:48:15 pm by MarkDS »
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #2 on: June 03, 2008, 09:52:57 pm »

Bill,

Further to the above, interesting that Rawanalyze shows the same kind of result we expected given what the histogram looked like in the camera LCD - no clipping. But then one wonders what Rawanalyse would do with Canon raw images? Do you think it has a generally more accurate fix on the raw data, or is this an anomoly affecting the treatment of the D3?
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Noise About Noise
« Reply #3 on: June 03, 2008, 11:06:22 pm »

Quote
Hi Bill,

Many thanks for reading and commenting. Appreciated. Some points:

First - something went wrong with the links to your illustrations - the first two started as the same image in Camera Raw 4.4., then as I was writing, the first two sorted out correctly. Weird.
[{POST_SNAPBACK}][/a]

I hit the wrong key and made my post prematurely, before I had refinished writing it. I then had to go back and edit it and had some trouble posting the images. So you were viewing my post as I was editing it. Sorry for the confusion.

Quote
Second, in the 3rd illustration is there any reason why you would prefer to reduce exposure rather than increase recovery to rescue the blown highlights?
[a href=\"index.php?act=findpost&pid=199639\"][{POST_SNAPBACK}][/a]

I used the exposure control since it is linear and moves the entire scale down by 0.5 EV whereas recovery is nonlinear and affects mainly the highlights and would not give the desired result.


Quote
Fourth, the kind of histogram you get in the raw converter of course depends on what the user decides to be the ACR defaults. I notice you have Brightness and Contrast set on the usual ACR defaults we download the program with. Perhaps in the defaults you are using there is also a moderate S in the Point Curve. These things can cause some highlight clipping when the highlights are very close to the edge in the raw file. I turn all that stuff off to zero in the basic tab and set the tone curves to Linear. Then I see things as they are without any "pumping-up" in the raw converter. It's what I like to think of as a "Greenfield" starting point. Then I build up. So with that background I can turn to your question: were the D3 shots properly exposed as ETTR?

The answer is a qualified YES, in the sense that when we made the D3 shots, on the camera's LCD screen the histograms looked bang-on ETTR, except for minor clipping of specular highlights. Downloaded into ACR with the settings at the flat defaults I discussed just above (and in footnote 6 in the article), lo and behold there was clipping of the brightest portion of the sky, which we did NOT see in the camera LCD. However, using ACR's Recovery function, the blueness of the sky got perfectly recovered. So I would say the D3 shots were ETTR+a bit.

I think perhaps one lesson of experience from doing this work is the usefulness of experimenting to select in-camera JPEG settings and raw converter default settings that will best produce similar histograms of the raw file in both places - should limit unpleasant surprises.
[a href=\"index.php?act=findpost&pid=199639\"][{POST_SNAPBACK}][/a]

I agree it is sometimes to set ACR to linear by setting contrast, brightness, and black to zero and using a linear tone curve. However, in this instance, it does not make that much difference in the rendering of the highlights as shown here:



In my own tests, I too have found that the camera histogram reflects the status of the raw histogram to within 0.3 EV, so your ETTR via the camera histogram is likely pretty close. Here is the camera histogram for the previously shown raw histogram. It shows slight clipping, which disappears with an exposure reduction of 0.3 EV.





I don't know how what Rawanalyze would show with the Canon, but it would be interesting to find out. Why don't you download it and see for yourself and report back to us.

[a href=\"http://www.cryptobola.com/PhotoBola/Rawnalyze.htm]Rawanalyze[/url]

Anyway, thanks for your excellent work. It will take some time to digest all the material.

Bill
Logged

JeffKohn

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1668
    • http://jeffk-photo.typepad.com
Noise About Noise
« Reply #4 on: June 03, 2008, 11:20:25 pm »

ACR does the same thing for my D300 files. I could not for the life of me figure out why my D300 NEF's look overexposed when I got them back to the computer even though the histogram looked like exactly what I wanted in-camera (I often use UniWB and linear tone curve in-camera).  I've since set my ACR defaults for the D300 to use -.5EV exposure adjustment and the images look as I expect. I know some have tried to explain in other threads why ACR's approach to 'normalizing' things makes sense, but in the case of the D3/D300 somebody screwed up.
Logged
Jeff Kohn
[url=http://ww

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #5 on: June 04, 2008, 07:31:42 am »

Quote
I don't know how what Rawanalyze would show with the Canon, but it would be interesting to find out. Why don't you download it and see for yourself and report back to us.


Anyway, thanks for your excellent work. It will take some time to digest all the material.

Bill
[a href=\"index.php?act=findpost&pid=199648\"][{POST_SNAPBACK}][/a]

Bill, thanks, appreciated.

Actually I do have Rawanalyze - but I haven't played with it yet. Your suggestion is a very good one, which I shall do; but it will have to wait a few days as I have a tight deadline on a piece of work related to the other part of my life. I'll report back here once I've had a chance to do it.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #6 on: June 04, 2008, 07:43:39 am »

Quote
ACR does the same thing for my D300 files. I could not for the life of me figure out why my D300 NEF's look overexposed when I got them back to the computer even though the histogram looked like exactly what I wanted in-camera (I often use UniWB and linear tone curve in-camera).  I've since set my ACR defaults for the D300 to use -.5EV exposure adjustment and the images look as I expect. I know some have tried to explain in other threads why ACR's approach to 'normalizing' things makes sense, but in the case of the D3/D300 somebody screwed up.
[a href=\"index.php?act=findpost&pid=199649\"][{POST_SNAPBACK}][/a]

Jeff, I'm not sure I would try to explain this in terms of somebody "screwing-up". We're talking about two indicators of exposure (and Bill refers to a 3rd, which makes a lot of sense). Each of those indicators is driven by all the assumptions that go into how they behave. So which do you want to use as the benchmark - the camera's histogram or the ACR histogram, and with what combination of default settings to render these histograms? The raw data of course is the raw data regardless of the in-camera JPEG settings which influence the camera's histogram, and ACR has suggested defaults (which we can change). The essential challenge here is to try to make them as coherent as possible in order to minimize unpleasant surprises. How one meters the scenes and what the camera does with that information should also influence these outcomes. I think this is less a matter of anyone screwing-up and more a question of alligning the variables to get results that are as close to coherent as possible and useful.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Noise About Noise
« Reply #7 on: June 04, 2008, 08:54:17 am »

Quote
Jeff, I'm not sure I would try to explain this in terms of somebody "screwing-up". We're talking about two indicators of exposure (and Bill refers to a 3rd, which makes a lot of sense). Each of those indicators is driven by all the assumptions that go into how they behave. So which do you want to use as the benchmark - the camera's histogram or the ACR histogram, and with what combination of default settings to render these histograms? The raw data of course is the raw data regardless of the in-camera JPEG settings which influence the camera's histogram, and ACR has suggested defaults (which we can change). The essential challenge here is to try to make them as coherent as possible in order to minimize unpleasant surprises. How one meters the scenes and what the camera does with that information should also influence these outcomes. I think this is less a matter of anyone screwing-up and more a question of alligning the variables to get results that are as close to coherent as possible and useful.
[{POST_SNAPBACK}][/a]

Mark,

My understanding of the ACR [a href=\"http://www.adobe.com/products/dng/pdfs/dng_spec.pdf]BaselineExposure [/url] compensation is to ensure that for a given exposure (say 1/60 s, f/8, ISO 1600), the resulting rendered images taken with various cameras will have the same appearance. Is this what you found in your testing with the Canon and Nikon cameras?

The issue revolves around how much headroom one should leave for specular highlights. The official ISO Specification for digital camera ISO speed ratings leaves 0.5 EV headroom for specular highlights and my own testing shows that the Nikon metering conforms to ISO standards.

However, for practical ETTR exposure it is often best to place non-specular highlights just short of clipping and let the specular highlights blow out. One can use negative exposure compensation in the raw converter to recover the specular highlights if desired. I understand that this is your approach and I agree that it is the best approach in most instances.

If my assumptions are correct, Canon does not leave this headroom for specular highlights.
« Last Edit: June 04, 2008, 08:57:18 am by bjanes »
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #8 on: June 04, 2008, 10:03:07 am »

Quote
Mark,

My understanding of the ACR BaselineExposure compensation is to ensure that for a given exposure (say 1/60 s, f/8, ISO 1600), the resulting rendered images taken with various cameras will have the same appearance. Is this what you found in your testing with the Canon and Nikon cameras?

The issue revolves around how much headroom one should leave for specular highlights. The official ISO Specification for digital camera ISO speed ratings leaves 0.5 EV headroom for specular highlights and my own testing shows that the Nikon metering conforms to ISO standards.

However, for practical ETTR exposure it is often best to place non-specular highlights just short of clipping and let the specular highlights blow out. One can use negative exposure compensation in the raw converter to recover the specular highlights if desired. I understand that this is your approach and I agree that it is the best approach in most instances.

If my assumptions are correct, Canon does not leave this headroom for specular highlights.
[a href=\"index.php?act=findpost&pid=199687\"][{POST_SNAPBACK}][/a]

Bill,

Regarding the exposures, the first idea was to focus on the shape and position of the in-camera histogram (whether the aim was an ETTL or an ETTR) and get that looking approximately the same between the three cameras for each ISO, whether or not it resulted in the same aperture and shutter speeds between them. So for ETTL the idea was to expose to just before the point of clipping blacks, and for ETTR, as you mention, the idea was to tolerate some specular highlight clipping. [But it turned out that a slightly more aggressive approach to ETTR ("ETFR" for the Canons) worked out better, after a bit of Recovery in ACR.]

It happened that the exposure compensation settings were quite different between the Canons and the Nkon to achieve this. Which takes us to the question of headroom. It looked very much to us that both Canons allowed more headroom for specular highlights in the sense that they average down the exposure in order to avoid clipping specular highlights - hence the midtones end-up further to the left than where ideally one would want them, whereas the Nikon produced a more central positioning of the midtones at the expense of some highlight clipping. These findings emerge from examining the histograms in ACR at the footnote 6 settings for that scene.

I think it would be good to revisit these histogram comparisons using Rawanalyze and see whether it confirms the initial observations. I intend to do that as soon as time permits. Meanwhile, your impression of my preferred exposure strategy is correct, except that once in ACR, whether to deal with clipping via implementing some Recovery or reducing the Exposure depends on the image: if the mid-tones look over-exposed at the same time the highlights are clipped, then of course it makes sense to reduce Exposure, but if the mid-tones look OK and there are clipped highlights one wishes to recover, then Recovery would be the tool of choice.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

joedecker

  • Full Member
  • ***
  • Offline Offline
  • Posts: 142
    • http://www.rockslidephoto.com
Noise About Noise
« Reply #9 on: June 04, 2008, 10:33:06 am »

Just a question--it's probably answered in the article, but I didn't see it.  In Figure 4, the article shows a comparison of tight crops from the 1Ds and 1Ds3, taken at 45mm and f/6.3 and f/7.1.  Now, those cameras have the same sensor size, and those images are taken at the same focal length, so my initial assumption as to what I was looking at in that picture must have been wrong--that is, I had assumed that the two pictures were quote-one-hundred-percent-pixel-crops-unquote.  Had that been true, the 1Ds3 image would have been larger in screen dimensions to show the same area of the scene.

So, there are at least four possible explanations I've come up with so far, but to entirely make sense of the resolution results it's important to know which is the case.

First, it's possible that that the 1Ds image is "100% pixels" and the other is downsampled.

Second, it's possible that the 1Ds3 image is "100% pixels" and the 1Ds is upsampled.

Third, it's possible that both images are resampled.

Fourth, it's possible that those images are both photographs of the print.

In any of these cases, I'd be curious what Figure 4 actually represents, and, by extension, what's happening in any of the other figures (it appears that a comparison of Figure 2 with Figure 3 leads to the same questions,  as an example.)
Logged
Joe Decker
Rock Slide Photography [url=h

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Noise About Noise
« Reply #10 on: June 04, 2008, 11:02:13 am »

Quote
I think it would be good to revisit these histogram comparisons using Rawanalyze and see whether it confirms the initial observations. I intend to do that as soon as time permits. Meanwhile, your impression of my preferred exposure strategy is correct, except that once in ACR, whether to deal with clipping via implementing some Recovery or reducing the Exposure depends on the image: if the mid-tones look over-exposed at the same time the highlights are clipped, then of course it makes sense to reduce Exposure, but if the mid-tones look OK and there are clipped highlights one wishes to recover, then Recovery would be the tool of choice.
[a href=\"index.php?act=findpost&pid=199694\"][{POST_SNAPBACK}][/a]

Mark,

We have discussed the use of the recovery and exposure sliders before and our approaches are different but not incompatible. In CR_09 of the ACR tutorial by Michael and Jeff Schewe, Thomas Knoll discusses the use of these tools. His preferred approach is to use the exposure slider to set the highlights short of clipping and then to use the brightness slider to set the mid tones. An alternate method, introduced in Lightroom, is to use the recovery slider to set the highlights and the exposure slider to set the mid tones. He states that the two methods are equally valid. Personally, I prefer the first method and you appear to prefer the second method.

To correct the BaselineExposure offset used by ACR with the D3, I think the proper tool is the exposure slider. The +0.5 EV offset used by ACR for the D3 moves the entire tonal scale (shadows, mid tones and highlights) 0.5 EV to the right and an exposure setting of -0.5 EV in ACR undoes this adjustment. Use of the recovery slider would move the highlights to the left, but the mid tones and shadows would still be too bright.

The situation in which I have found the Recovery slider to be most useful is with backlit and sidelit shots where the overall exposure is correct but the highlights are burnt out.
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #11 on: June 04, 2008, 05:04:46 pm »

Quote
Just a question--it's probably answered in the article, but I didn't see it.  In Figure 4, the article shows a comparison of tight crops from the 1Ds and 1Ds3, taken at 45mm and f/6.3 and f/7.1.  Now, those cameras have the same sensor size, and those images are taken at the same focal length, so my initial assumption as to what I was looking at in that picture must have been wrong--that is, I had assumed that the two pictures were quote-one-hundred-percent-pixel-crops-unquote.  Had that been true, the 1Ds3 image would have been larger in screen dimensions to show the same area of the scene.

So, there are at least four possible explanations I've come up with so far, but to entirely make sense of the resolution results it's important to know which is the case.

First, it's possible that that the 1Ds image is "100% pixels" and the other is downsampled.

Second, it's possible that the 1Ds3 image is "100% pixels" and the 1Ds is upsampled.

Third, it's possible that both images are resampled.

Fourth, it's possible that those images are both photographs of the print.

In any of these cases, I'd be curious what Figure 4 actually represents, and, by extension, what's happening in any of the other figures (it appears that a comparison of Figure 2 with Figure 3 leads to the same questions,  as an example.)
[a href=\"index.php?act=findpost&pid=199699\"][{POST_SNAPBACK}][/a]

None of the above   .

The dimensions (width and height) of a photograph can be altered by entering the Image Size dialog, unchecking the resample "Resample" box, and entering the W and H one wants. The PPI will vary inversely with the change in dimensions but there will no resampling. As long as PPI is 180 or above it suffices for the purposes of this article.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #12 on: June 04, 2008, 05:12:30 pm »

Quote
Mark,

He states that the two methods are equally valid. Personally, I prefer the first method and you appear to prefer the second method.

[a href=\"index.php?act=findpost&pid=199701\"][{POST_SNAPBACK}][/a]

Bill, no problem with what you're syaing, but the approach I prefer depends entirely on what I think the image needs. So, over-exposed highlights, decent mid-tones and shadows: I use Recovery. Over-exposed everything: I reduce Exposure; Under-exposed shadows, over-exposed highlights but decent mid-tones: I leave Exposure alone and use Fill+Blacks and Recovery. I.O.W. "it all depends" - basic principle - target the adjustment to where the problem is. That's what makes the latest incarnations of ACR such valuable image editing tools - I'm sure you would agree; lots of flexibility to do many things in a more targeted way than how it was in 3.7 and earlier.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Nick Rains

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 705
    • http://www.nickrains.com
Noise About Noise
« Reply #13 on: June 04, 2008, 06:03:01 pm »

Quote
In my own tests, I too have found that the camera histogram reflects the status of the raw histogram to within 0.3 EV, so your ETTR via the camera histogram is likely pretty close. Here is the camera histogram for the previously shown raw histogram. It shows slight clipping, which disappears with an exposure reduction of 0.3 EV.
[a href=\"index.php?act=findpost&pid=199648\"][{POST_SNAPBACK}][/a]

Are the camera histograms based on the RAW file or the camera's jpeg preview?

Since the image displayed on the camera screen is a jpeg preview not a RAW image am I correct in assuming that the histogram is derived from that data, not the actual RAW data.
Logged
Nick Rains
Australian Photographer Leica

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #14 on: June 04, 2008, 06:28:08 pm »

Quote
Are the camera histograms based on the RAW file or the camera's jpeg preview?

Since the image displayed on the camera screen is a jpeg preview not a RAW image am I correct in assuming that the histogram is derived from that data, not the actual RAW data.
[a href=\"index.php?act=findpost&pid=199754\"][{POST_SNAPBACK}][/a]

Nick, this is exactly the issue - the histograms on the camera screen are generated from the JPEG settings.

Mark
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

bernie west

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 323
    • Wild Photo Australia
Noise About Noise
« Reply #15 on: June 04, 2008, 09:10:58 pm »

Quote
Nick, this is exactly the issue - the histograms on the camera screen are generated from the JPEG settings.

Mark
[a href=\"index.php?act=findpost&pid=199760\"][{POST_SNAPBACK}][/a]
The other issue is that the in-camera histogram (at least for the non pro cameras, don't know about the 1 series) is some undefined (as far as i know) averaging of the three channels.  Probably mostly influenced by the green channel.  So you can get clipping in one or two channels, which will show as clipping in your raw converter, but as long as the 'averaged' value is not clipping, the in-camera histogram will not show clipping.  It would be good to be able to configure through firmware how the in-camera histogram actually performs.
Logged

Ray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10365
Noise About Noise
« Reply #16 on: June 04, 2008, 11:39:25 pm »

My 5D Picture Style jpeg (and therefore histogram) is set to minimum contrast, minimum sharpening and minimum saturation. The image review on the camera's LCD panel doesn't look too impressive when I sometimes decide to show someone the shot I've just taken.

On one occasion, a young fella after seeing a few reviews of some recent shots on my 5D's LCD, pulled out his iPhone with built-in 3mp camera and 3" screen and proceeded to display a number of sparkling and colorful shots which he reckoned were much better   .

I understand that ACR will try to reconstruct a blown channel from information contained in the other two channels. I don't know how this is possible or to what extent it can be done. However, if a processed image with approprite -EC adjustment does not appear to have any clipped channels, then for most practical purposes it doesn't have any clipped channels. The fact that a program like Rawanalyze can demonstrate that a particular channel is actually clipped is only of academic interest.

When attempting to maximise the camera's S/N by pushing exposure to its maximum, just short of clipping, one inevitably gets a few shots that are truely and irredeemably clipped, due to miscalculation. Common examples of obvious clipping in my own situation are blue skies with a distinct shift towards cyan and reddish skin tones that turn grey.

I recall some years ago seeing an elaborate technique on LL describing a method of reconstructing such blown channels. I decided it was much easier to be more careful at the shooting stage and bracket exposure when in doubt   .
Logged

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #17 on: June 05, 2008, 08:36:01 am »

Quote
It would be good to be able to configure through firmware how the in-camera histogram actually performs.
[a href=\"index.php?act=findpost&pid=199785\"][{POST_SNAPBACK}][/a]

To some extent you can with whatever JPEG settings the camera firmware provides. Ray's post gets into that.
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

Mark D Segal

  • Contributor
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 12512
    • http://www.markdsegal.com
Noise About Noise
« Reply #18 on: June 05, 2008, 08:47:50 am »

Quote
I understand that ACR will try to reconstruct a blown channel from information contained in the other two channels. I don't know how this is possible or to what extent it can be done.

When attempting to maximise the camera's S/N by pushing exposure to its maximum, just short of clipping, one inevitably gets a few shots that are truely and irredeemably clipped, due to miscalculation.

I decided it was much easier to be more careful at the shooting stage and bracket exposure when in doubt   .
[a href=\"index.php?act=findpost&pid=199803\"][{POST_SNAPBACK}][/a]

Ray,

ACR has a great Recovery function. Needless to say I don't know the math - I suppose it would haveto be some form of borrowing from unclipped data. What I do know, however, is that very often it is really, really effective. It does liberate the photographer from being overly-concerned about minor amounts of clipping. I agree with you that in attempting to maximise an ETTR one can miscalculate and over-do it, but depending on how one has the in-camera JPEG configured, at least with the Canons I've been using one is usually warned (after the fact) and can then adjust and retake - as long as it's not one of those "fleeting moment" images. When in doubt of course bracketing is good practice, but with a !ds3 and its 25~30 MB raw files it fills cards QUICKLY!
Logged
Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....."

bjanes

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3387
Noise About Noise
« Reply #19 on: June 05, 2008, 10:29:57 am »

Quote
The other issue is that the in-camera histogram (at least for the non pro cameras, don't know about the 1 series) is some undefined (as far as i know) averaging of the three channels.  Probably mostly influenced by the green channel.  So you can get clipping in one or two channels, which will show as clipping in your raw converter, but as long as the 'averaged' value is not clipping, the in-camera histogram will not show clipping.  It would be good to be able to configure through firmware how the in-camera histogram actually performs.
[{POST_SNAPBACK}][/a]

Bernie raises a good point about in camera histograms. They are said without much proof to represent luminance histograms where the RGB channels are weighted. [a href=\"http://www.cambridgeincolour.com/tutorials/histograms2.htm]Sean McHugh[/url] has a good tutorial in his Cambridge in Color site, where he discusses luminance and RGB histograms. The RGB weighting factors he gives are 59%, 30%, 11% for GRB respectively. Since the factor for blue is only 11%, luminance histograms are not sensitive for detecting clipped blues.

Most higher end cameras have an option to display separate histograms for the RGB channels. These are similar to the histogram one sees in ACR and by default in Photoshop (PS has an option to display luminance histograms). These RGB camera histograms are white balanced and do not show the true status of the RGB channels. To achieve white balance the red and blue channel values are multiplied by a coefficient greater than unity and the green channel is left alone.

Julia Borg has published these coefficients for some Nikon cameras according to the color temperature of the illuminant. For daylight, the red channel is most underexposed and needs the largest multiplier, while the blue is most underexposed with tungsten illumination. For the Nikon D3 with daylight, the RGB multipliers are about 2.2, 1.0 and 1.19. Shown below is a typical raw histogram for the D3 with daylight WB:



As is evident from the histogram, the red channel has the greatest headroom, followed by the blue channel. The bold tic marks are full f/stops and the other marks indicate thirds of a stop. In this case, proper ETTR was not achieved since the green channel is considerably below clipping, about 2/3 of an f/stop. With increased exposure, the green channel would start to clip, but the other channels would still be intact and highlight recovery in ACR would be possible. With still more increased exposure, the blue channel would start to clip, followed by the red channel. When all channels are clipped, highlight recovery in ACR is not possible. The red histogram is about 1 f/stop below the green, which indicates that a maximum of about 1 stop of highlight recovery is possible with this camera and daylight illumination.

With ACR at default settings, the above image appeared slightly overexposed according to the ACR histogram, due in large part to the +0.5 EV baseline exposure compensation that ACR uses for the D3. This demonstrates the importance of checking the raw histogram, which is easily done with Rawanalyze. If one wants the camera RGB histogram to give a better picture of the status of the RGB channels, one can upload to the camera a white balance where the WB multipliers are all unity (1.0). Search UniWB for details.
« Last Edit: June 05, 2008, 10:32:32 am by bjanes »
Logged
Pages: [1] 2 3   Go Up