Processing RAW images for accurate color measurements in scientific applications

[efisher16]

Hi all,

I'm new here, but very happy to have joined CiC. I have a question for everyone about how to correct the color cast and curves within a RAW photo so that accurate colors in RGB color space can be measured. I am doing some scientific research where color measurement from photos is very important (geological strata, archaeological artifact colors, etc). The photos are not for print, thus I am not using CMYK color space. I also don’t need the measurements to be insanely accurate, like to 1 or 2 RGB points, but I would like to get it as close as possible, and certainly to the point where there are negligible visible differences between the images and 'real-life' colorimeter measurements. It would be nice to even be able to quantify the range of variation between the colors in an image compared to real-life standards so that I can discuss how well the image colors match.

Currently, I am taking my photos under consistent lighting conditions. My predominant light source is a Nikon D800 strobe. Due to field conditions I can never restrict all other light sources, so there is always some component of background sun light, but I try to minimize it. I am using a D300s with a Nikon Nikkor DX 18-200 mm f3.5-5.6GII lens. The white balance is set to flash so that it remains a constant 5500 degrees. I place white, black, and 18% grey cards from Digital Image Flow in the scene, which I am able to use later in Photoshop. The white and black cards help me to adjust exposure while the grey card adjusts for the color cast. After I download my photos I use a personalized lens calibration file specific to my lens and camera to minimize geometric distortion, vignetting, and chromatic aberration. My monitors are calibrated using a Spyder Pro.

Lately, I have been taking photographs of a Munsell color chart, correcting my photos, and then comparing the published RGB values of the actual Munsell colors against my image results. The DGK grey card seems to very accurately remove color cast and my RGB values only vary by 1 or 2 points. The problem is that the entire curve is typically skewed positive or negative, which means that the colors are accurate relative to each other in the photos but not accurate to absolute colors. I have been toying with using the exposure slider (but not brightness! great CiC post, by the way) to shift the entire curv up or down so that my measured RGB value of the grey card matches its actual measured values (i.e. real life). This method produces decent results and my color comparisons with the Munsell colors are all within ~20-30 RGB points (mean = 17, sd = 12.4), which is a barely perceptible difference in hue to the human eye on a neutral background. However, by adjusting the brightness of the image it displaces the whites and blacks.

I keep thinking though that there must be some better method to accurately correct the colors in photos? Is there a way to correct each band (R, G, and B) individually? Does anyone have any experience with this or know of any papers, etc where some methods are described? I would really appreciate any feedback.

Best,

Erich

[herbert]

Hi Erich,

You can create a custom colour profile for use in Camera RAW. Basically you take a photo of a known set of colours. This can be used to build a colour profile that is applied to all your images shot under the same light.

Unfortunately you have to buy even more gear. See the following for details of the X-rite color passport:

http://www.northlight-images.co.uk/r...assport_1.html
http://www.luminous-landscape.com/re...psssport.shtml

Hope this helps.

Alex

[Manfred M]

ACR and Nikon do not go together very well. A number of Nikon shooters that I know, including serveral on this site have found the same issue and they will use View NX2 to Capture NX2 for the RAW conversion and save to TIFF and then edit the TIFF file in Photoshop.

[Simon Garrett]

ACR and Nikon do not go together very well. A number of Nikon shooters that I know, including serveral on this site have found the same issue and they will use View NX2 to Capture NX2 for the RAW conversion and save to TIFF and then edit the TIFF file in Photoshop.

I know some people say that, but if I may say so I think it's a matter of opinion! I find LR4 is generally better than Nikon Capture NX2 at rendering my D300 raw images.

NX2 rendering is closer to the in-camera jpegs, obviously, as the software is designed with that intention. That doesn't IMHO make it better - just different. The latest Adobe "Camera" profiles for Nikon cameras (or for the D300, at any rate) are very close in colour rendition to the Nikon Picture Controls, but tone mapping and noise control in LR4.2 are (again, IMHO) far better than can be achieved in NX2.

I mean, what would you expect? Nikon Capture NX2 is more than 4 years old, and has advanced in features and image rendering very little in that time. In the same time, Lightroom has gone through 3 versions (2 to 4) with major enhancements to the rendering engine.

However, one note of caution for the OP: in normal photography, accurate scene-referred colour rendition is not what's wanted. For record photography then you probably do want that, and I'm not surprised at the problems you're finding. You might want to search scientific subject-related sources for advice on scene-referred colour rendition, rather than asking "normal" photographers!

[revi]

Erich,

If I understand correctly, you measure your final rgb values in the developed picture, so the screen/display colours don't play a role?
That would remove one big source of errors.

The first thing to keep in mind is that the black and white cards you use are neither black nor white, meaning that the black isn't at rgb(0,0,0)
and the white isn't at rgb(255,255,255). This should be expected, as you cannot exceed the dynamic range of a paper print, which is about 4 stops.
Your camera sensor can easily cover at least 10 stops of dynamic range.

If you take shots of the Munsell chart at the same time as those of your sample (and under the same light!), colour correction based on
the chart should be reliable. If you take the shots of the chart under different conditions, you can still get good results (given the black,
white and gray references you have), but perhaps less reliable.

Edit:
After looking a bit for the Munsell chart that was mentioned, I get the impression that it's not a chart intended for profiling of equipment. But as long
as you have reliable reference values, that shouldn't be a problem.

[John Morton]

ACR and Nikon do not go together very well. A number of Nikon shooters that I know, including serveral on this site have found the same issue and they will use View NX2 to Capture NX2 for the RAW conversion and save to TIFF and then edit the TIFF file in Photoshop.

I'm one of the people who convert my Nikon NEF RAW files to TIFFs in Capture NX2, then edit in Photoshop.

I don't know if this would be helpful to you; it is a plug in filter from Power Retouche for 'White Balance':

http://powerretouche.com/White-balan...n_tutorial.htm

You will note that one of the options it offers is:

"Gretag Macbeth Color Rendition Chart is a standard color reference chart used by photographers. PowerRetouche has expanded the charts use for digital photography by implementing white balance correction from the six gray patches. You do not have to pick all six."

I've actually never used that particular option before, but I thought I'd give it a try now. So, here's the image I had already 'finished' with:



...and here is the same image after using all six grey patches on the xRite Color Checker target:



I just used a JPEG that I had handy here on my Internet computer.

Hmm... I think I'll have to start using that 'White Balance' option a little more often!

[Manfred M]

I know some people say that, but if I may say so I think it's a matter of opinion! I find LR4 is generally better than Nikon Capture NX2 at rendering my D300 raw images.

NX2 rendering is closer to the in-camera jpegs, obviously, as the software is designed with that intention. That doesn't IMHO make it better - just different. The latest Adobe "Camera" profiles for Nikon cameras (or for the D300, at any rate) are very close in colour rendition to the Nikon Picture Controls, but tone mapping and noise control in LR4.2 are (again, IMHO) far better than can be achieved in NX2.

I mean, what would you expect? Nikon Capture NX2 is more than 4 years old, and has advanced in features and image rendering very little in that time. In the same time, Lightroom has gone through 3 versions (2 to 4) with major enhancements to the rendering engine.

However, one note of caution for the OP: in normal photography, accurate scene-referred colour rendition is not what's wanted. For record photography then you probably do want that, and I'm not surprised at the problems you're finding. You might want to search scientific subject-related sources for advice on scene-referred colour rendition, rather than asking "normal" photographers!

Simon - I don't remember suggesting that the Nikon software was better for working the images, just for converting them into a form that bypasses the Adobe RAW engine, that seems to be the source of the problem.

There was a posting on the Adobe website a few months ago, that seems to have been removed, that was all about how both Adobe and Nikon wanted the best for photographers. That certainly raised suspicions that perhaps that the two companies were perhaps not working as closely as Nikon users would like.

My suspicion is that the Adobe RAW engine is quite generic as it has to handle input from various camera manufacturers and camera models to convert the images. This implies a mapping table approach, where each camera / model needs a separate one. If the cooperation between Nikon and Adobe is not as close as it should be, then Adobe would have reverse engineered the Nikon data to run through their generic RAW engine. This means that the output would be close, but certainly not completely correct.

Your comments on the software are not quite correct; while the base package has been around for a while, Nikon does release updates as new camera models are released. I had to wait for about a month after I got my D800 before I could use RAW files. It took about the same length of time for Adobe ACR to be released and DxO took even longer.

For most of my work, ACR is close enough, but I have run into a number of cases where I could not get the skin tone to look right without having some other colours in the image not look quite right. I just wish that Adobe and Nikon would stop playing corporate games and would fix the problem.

[Colin Southern]

My suspicion is that the Adobe RAW engine is quite generic as it has to handle input from various camera manufacturers and camera models to convert the images. This implies a mapping table approach, where each camera / model needs a separate one.

And indeed each camera/model does indeed get one.

If the cooperation between Nikon and Adobe is not as close as it should be, then Adobe would have reverse engineered the Nikon data to run through their generic RAW engine.

No, not at all; it's only the final result that matters, and that can be achieved simply by comparing that data returned by the camera from a known target. No reverse engineering required.

[Manfred M]

No, not at all; it's only the final result that matters, and that can be achieved simply by comparing that data returned by the camera from a known target. No reverse engineering required.

Colin - I know you stick to that line, but from a technical standpoint, I don't buy it. What you are saying will only work if the Adobe and Nkon (etc.) RAW processing engines work the same way. A raw file is simply a string of pixel based data elements with a bit of additional information in the header file. Assembling that data into coherent RGB files that can be edited is something that the RAW engine does. If the way that the the Adobe and Nikon engines build the image is different, for instance the weighting of non-adjacent pixels, the the conversion results would be different. This is really the reverse engineering I am referring to.

The problem with reverse engineering (and I've done enough of it to understand the upsides and downsides) is especially acute if you want to fit the results into an existing data model. That forces you to massage the data to fit your model, even if it is not quite as accurate as you would like. Building a new data model can give better results, but then you have to go through the expense of building and testing the new model and migrating all of your legacy data. This isn't clean either, because you could end up breaking the legacy data. You could try to maintain two data models, but that is just asking for trouble.

Comparing the data that comes from a known target will let you calibrate a model, assuming that a single data point is good enough, but it will not get around underlying differences in how two different processing engines work.

[Colin Southern]

Colin - I know you stick to that line, but from a technical standpoint, I don't buy it. What you are saying will only work if the Adobe and Nkon (etc.) RAW processing engines work the same way. A raw file is simply a string of pixel based data elements with a bit of additional information in the header file. Assembling that data into coherent RGB files that can be edited is something that the RAW engine does. If the way that the the Adobe and Nikon engines build the image is different, for instance the weighting of non-adjacent pixels, the the conversion results would be different. This is really the reverse engineering I am referring to.

The problem with reverse engineering (and I've done enough of it to understand the upsides and downsides) is especially acute if you want to fit the results into an existing data model. That forces you to massage the data to fit your model, even if it is not quite as accurate as you would like. Building a new data model can give better results, but then you have to go through the expense of building and testing the new model and migrating all of your legacy data. This isn't clean either, because you could end up breaking the legacy data. You could try to maintain two data models, but that is just asking for trouble.

Comparing the data that comes from a known target will let you calibrate a model, assuming that a single data point is good enough, but it will not get around underlying differences in how two different processing engines work.

What you're referring to is primarily the demosaicing process and in that repect - yes - ACR does it the same way for all makes and models (with the colorimetric interpretation data contained on the decoder ring metadata being the main variable). What do Nikon do that's "better" and that Adobe would want to reverse engineer and copy? Nothing that I can think of; at the end of the day, be it from a Nikon or Canon or any other make, it's still just rows and columns of RGB information ... nothing secret about that.

What makes a far bigger difference is in how that information is Gamma encoded to match the typical response of the human eye ... and that's something that's a core function of the product, not something that needs to be reverse-engineered for certain makes and/or models.

Essentially, what comes out of any camera is normalised data; the only variable is the colorimetric interpretation metadata, which they weight and/or validate with their own profiles.

[Simon Garrett]

What you are saying will only work if the Adobe and Nkon (etc.) RAW processing engines work the same way.

To add to Colin's reply, what does Nikon know (that Adobe doesn't) that is pertinent to raw conversion?

Not just "they made the sensor so they know all about it", but what specific information that makes Nikon raw conversion better than Adobe's?

[Manfred M]

The information I read some time ago was that Nikon did not simply uses the 2x2 data (R-G-B-G) to generate the colour information for an individual display pixel, but used a weighted algorithm to determine the colour from other nearby photodiodes. Unless the algorithm is known, accurate decoding is not possible. Don't forget in NEF format, the data is transfered at a photodetector, not pixel level and the RAW engine has to assemble the data at that level.

The fact that the Nikon supplied software seems to do a better job (colour accuracy) than ACR suggests that there is a problem with the Adobe RAW engine.

[Colin Southern]

The fact that the Nikon supplied software seems to do a better job (colour accuracy) than ACR suggests that there is a problem with the Adobe RAW engine.

Argh!

Better than ACR with ACR set to which profile exactly? (because there are many and they ALL produce different colours).

I know you like NX2 (as do a lot of other folks), but one just can't say "it's more accurate than ACR" when 99% of the accuracy of ACR colour comes from the specific profile that its applying to the RAW data.

[Manfred M]

Argh!

Better than ACR with ACR set to which profile exactly? (because there are many and they ALL produce different colours).

I know you like NX2 (as do a lot of other folks), but one just can't say "it's more accurate than ACR" when 99% of the accuracy of ACR colour comes from the specific profile that its applying to the RAW data.

Colin – actually I think that the variants of NX2 are terrible to use, but they do one thing very well; accurate Nikon RAW conversions. can’t think of too many people that sing its praises for any other functionality.

I look at DxO in much the same light, not a great piece of software to use, but it really corrects lens defects very well. I have tested that software against ACR lens / camera corrections and find it does a much better job. It also covers a lot of camera lens combinations that Adobe does not.

I have a lot of design and engineering background (including around a decade of software development), so I strongly suspect that the real culprit is the design of the Adobe RAW engine. If neither the Nikon nor Adobe product provided good results, I would have drawn a different conclusion, but seeing as one of the products works right out of the box, that indicates that there is a problem with the other product.

I think I have found an example that clearly shows that ACR does not handle Nikon files properly and one is outside of the profiling discussion.

I posted an image of the red sand dunes near Sossusvlei, in the Namib Desert and was trying to figure out why I had a white line between the top of the dune and the sky. The image was taken during “Golden Hour” just before sunset, so I initially thought it might be an optical effect.


The Giant Sand Dunes at Sossusvlei at Sunset




I went back and opened the RAW file in ACR and converted to a jpeg, which I posted. This image is just as I opened it and then saved it as a jpg. All profiles were turned off.



I then took a sample of a closeup of the white line.



I discovered that the image I had posted came from an ACR processed RAW file, so I went back I then repeated the exercise, this time using View NX2.





While the colours are more subtle and I think more accurate in the NX2 version, and I suppose that could be explained with the profile or ACR engine issue. However but the lack of artifacts, i.e. no white line in the Nikon RAW engine output, versus a very distinct artifact introduced by the Adobe RAW engine makes me suspect that the problem is a result of the software. I did play around in ACR to try to get rid of the line.

This line really pops out once I start adjusting the levels to make them closer to the sunset view that it was.

Any thoughts?

[Colin Southern]

All profiles were turned off.

Hi Manfred,

At the risk of "flogging a dead horse", I do have to wonder if you're missing a "puzzle piece" when it comes to profiles (but I don't mean that in any nasty way).

Basically, there's NO WAY to turn profiles off in ACR - the conversion has to be based on SOMETHING. If it's not a custom camera profile created by the likes of Color Passport - then it may be Adobe Standard or Camera Neutral or Camera Faithful - but - it has to be something and that "something" WILL have a significant effect on colour, including colour gradients.

In the example you quoted previously where you said that you believe that NX2 may base the demosaicing on something other than RGBG co-sampling, it may make a small difference to how the colour of fine detail is resolved, but in the vast majority of cases, images displayed on a typical monitor (ie a full-resolution image displayed at 25%) is going to be severely down-sampled, and in all likelihood consist of filled shapes that are of such a size that any colour influence from a RGBG co-sampling or some more exotic algorithm is going to be long sampled out. In fact the only thing I can think of that Nikon might be able to account for that Adobe probably couldn't, would be sensor non-linearity, but again, I think that a custom camera profile is still going to do the better job because it'll also be compensating for differences in lens colour, spiky lighting etc.

What I'd REALLY love to see is a measured comparison between a 24 patch colour chart shot on a Nikon and processed with both NX2 and ACR (with ACR using an accurate colour profile for that camera / lens / lighting combination). I'd love to do it myself, but sadly I don't have a Nikon nor NX2.

For what it's worth, folks say the same thing about Canon software and ACR with respect to colour ... and all I can say in reply is that I use ACR with custom profiles and have very little problem with colour -- and the problems I do have are more metamerism related, that any software would struggle with.

[Manfred M]

You are quite right Colin; I should not have said “no profile”, but rather “default profile”. I do know that because the software does need a set of rules to apply in the absence of a profile having been selected. I would not expect identical results out of the two sets of software, as Adobe and Nikon are likely to have a slightly different view of what the different profiles should be, even though they tend to give them the same names. Regardless, the choice was made to try to minimize significant processing that might introduce any artifacts.

I also know about the issues related to downsampling. That is why I included the two full-size images and two detailed images, showing the differences between ACR and NX2. The detail images were both shown at 100% for that reason. Sorry that I did not explicitly state that in my posting. What I was trying to demonstrate is that the two pieces of software do give completely different results when in default mode. While colour accuracy is more difficult to argue without a shot with a test swatch, the artifacts from the conversion process are harder to ignore.

I haven’t done any side-by-side controlled tests for colour. I have an x-rite mini color checker and could look at trying that once I’m mobile again (a few more months unfortunately).

Metamerism is something that is a problem with no real solution. I tend to worry about skin tones and get less excited about material colours matching exactly. I suspect it is more of a concern for you as a professional photographer.

I worked I the garment industry many years ago and that was always a problem when buying from different suppliers. We used a Macbeth light booth for incoming quality checks to make sure that the various dye lots were in spec. Customers would buy something in a mall with the product having been tried on in a no-daylight, tungsten / fluorescent lighting situation and would then notice a colour problem when they got home and would complain. Thankfully, I ran the engineering department, not quality control or customer relations, so those problems were someone else’s to solve.

Back to the matter at hand, I still suspect that the ACR RAW engine is the problem, whether that is because of a lack of cooperation between Adobe and Nikon or if the underlying design is at issue, Regardless, the NX2 work flow seems to be the one I am going to stick with,

[Colin Southern]

You are quite right Colin; I should not have said “no profile”, but rather “default profile”. I do know that because the software does need a set of rules to apply in the absence of a profile having been selected. I would not expect identical results out of the two sets of software, as Adobe and Nikon are likely to have a slightly different view of what the different profiles should be, even though they tend to give them the same names. Regardless, the choice was made to try to minimize significant processing that might introduce any artifacts.

OK - got that bit. So the next thing is "why would anyone who is serious about colour management be using the default profile for ACR (or NX2 for that matter)"? It's no better than using a manufacturers monitor profile or printer manufacturers paper profile.

While colour accuracy is more difficult to argue without a shot with a test swatch, the artifacts from the conversion process are harder to ignore.

That to me looks simply like a bit of sharpening has been applied. Is ACR set to apply sharpening to previews only?

Metamerism is something that is a problem with no real solution. I tend to worry about skin tones and get less excited about material colours matching exactly. I suspect it is more of a concern for you as a professional photographer.

Yeah - after spending $2k on makeup and discovering that I might be getting metamerism under the studio lights, I was just "happy happy happy"

[Simon Garrett]

Any thoughts?

Yes - could you post the original NEF file please, so we can have a go in both NX2 and Lightroom/ACR please?

Just to add: I've tried a few of my recent photos in Lightroom 4.2 (=ACR7.2) where there's a similar sky/land boundary, and can see no similar line artefact.

Using Landscape profile in Lightroom (or Landscape picture control in NX2) there is virtually no difference in colour rendition between Nikon (Capture NX2) and Adobe (Lightroom 4.2) with the pictures I've tried. I agree with Colin that comparing with Adobe's default profile is certainly going to show a difference in colour, as it's not designed to match Nikon's profile.

Edited to add: if I switch to Adobe's "Process Version 2003" (that's 2 ago - we've had PV2010 and PV2012 since), and also using the ACR 4.3 profile, I can get a slight line artefact on high-contrast edges at default settings. It looks to me that default sharpening in PV2003 results in a bit of over-shoot. I don't see that on PV2012.

[xpatUSA]

Hi all,

I'm new here, but very happy to have joined CiC. I have a question for everyone about how to correct the color cast and curves within a RAW photo so that accurate colors in RGB color space can be measured.

I keep thinking though that there must be some better method to accurately correct the colors in photos? Is there a way to correct each band (R, G, and B) individually? Does anyone have any experience with this or know of any papers, etc where some methods are described? I would really appreciate any feedback.

Best,

Erich

Hi Erich,

If you download some free software from Adobe you can do exactly what you're asking about but you'll also need a Gretag-Macbeth color-checking card (just the card, not Passport).

Assuming you already have Adobe Camera Raw: You need Adobe's RAW to DNG converter and also their DNG Profile Editor. Briefly, you shoot the card under your lighting conditions as well as the subject preferably, but not necessarily, in the same shot. You convert the shot with the card in it to a DNG file. You open that file in the profile Editor and go to the page where you select the corner patches of the card and, bingo, a profile is created complete with corrected white balance. You get many, many sliders to play with, not that you should need to. You save this profile to wherever Abode keeps them on your computer. You re-open the original RAW file with the subject in it using ACR, your new profile, and set any sliders that affect color to zero (saturation, vividness, etc). If you stay in ACR, the color picker should show you the RGB values. I have ACR 5.4. Later versions could show you hue, saturation and brightness/lightness which is what you are looking for.

Remember that the colors you see on-screen is a conversion of the "working color space" to sRBG, so don't use a screen color picker to measure the subject colors. In Adobe ACR 5.4/ PSE6 the color picker gives an 8-bit version of the 16-bit working color space, whereas a screen color picker can give quite different values, trusssst me ;-)

Also beware of subject colors that are out-of-gamut with respect to sRGB and Adobe RGB too. They will get changed in the output file, including what you see on-screen. If possible, work in what is called ROMM or probably ProPhoto in your editor.

Of course you already know that HSB or HSL are the best numbers to determine color, or L*a*b* if you want to get fancy. In other words, comparing the RGB from your file with the RGB table provided with the color card will be confusing. Hint: change the exposure and all three RGB numbers will change . . .

Good Luck!

[Simon Garrett]

Remember that the colors you see on-screen is a conversion of the "working color space" to sRBG

Just a query - why would the on-screen colour be sRGB? Surely it would be a conversion of the working colour space to the monitor colour space (as specified in the monitor profile)? Or are you saying that by way of simplification, as most monitors have a colour space of approximately sRGB?