Camera Profiling - DNG, ICC and alternative methods

[Vladimirovich]

how useful would the lower one be if I didn't like the color?

not useful indeed... that is the point - for as long as the words "I built a better profile" means "I built some profile that I subjectively like more than Adobe's" there is nothing to argue about... I am not arguing about tastes, but I suspect that OP has something else in mind and if so - let him come up w/ the procedure to test that will yield some dEs vs something instead of tastes and opinions

[Vladimirovich]

I am sure he can

he just wrote that it was subjective, sorry

[Vladimirovich]

Wondering what additional capability does more hueDivisions and satDivisions have, and what is lost with less valDivisions. Eric mentioned that the DNG PE cannot make lightness corrections at the moment.

the best capability is actually the ability of ACR/LR to work with just matrix profiles... + linear tone curve +baseline exposure = 0... and I trust Adobe to come w/ good matrices... the rest I can do to my taste (using WB target if necessary) w/o shooting any color targets (only may be as an exercise sometimes for a very very odd light... but I had recently for a colored w/ magenta gel light and even there I ended up, thanks to Eric C. for reminding about RTFM  and how it is done for IR modified cameras,  just adjusting WB range for a standard profile /with LUTs removed/ using Adobe PE - ended w/ absolutely no need to use any color targets).

[samueljohnchia]

Look, I have no ulterior motive in all this. I am just another natural light photographer chasing elusive perfection. If we realize that in all this camera profiling is a complete farce, I will be happy to swallow that. Right now I am honestly a bit confused as to how we are going to approach this testing. While I would very much like to know how all this works under the hood (yeah right, as if anyone is going to tell), my eyes are telling me the opposite at the moment. And for florescent lights the Adobe Standard profile seriously sucks.

Andrew, I am sure that with reference values from various color targets, we can evaluate the accuracy of the profiles by comparing with rendered values of photographed targets? But scene colors that are not on a target are going to be harder

[digitaldog]

And for florescent lights the Adobe Standard profile seriously sucks.

How about a custom DNG profile, does it suck?

If no, that alone is telling in terms of the possible testing.

Now, if you had a converter that allowed DNG and ICC profiles, which would you find better (easier, faster)?

[samueljohnchia]

How about a custom DNG profile, does it suck?

I set up a simple scene of various red-orange objects at my computer desk. I made a custom DNG profile under the same illumination. The custom DNG profile was shockingly close to the real deal. It was a simple matter of adjusting the WB sliders by eye to get a close match.

The standard untwisted profile was very difficult to adjust to the real scene. I could not get a close match.

Now, if you had a converter that allowed DNG and ICC profiles, which would you find better (easier, faster)?

Good question. I will try to make an ICC profile with a CC and run in Raw Therapee. At present, with DNG profiles, RawTherapee does not seem to be able to utilise dual illuminant tables. I am also not sure of how it utilizes the huesatDelta tables. I say this because I can get RT to practically the exact same colors as ACR, except with a slight warm tint to the colors. Note that the neutrals are not warm tinted, only the colors.

[samueljohnchia]

This is another excellent post from Eric on optimizing color transforms in camera profiles: http://www.luminous-landscape.com/forum/index.php?topic=37682.msg326214#msg326214

I am now highly doubtful of an approach to compare photographed chart colors to reference synthetic targets as a sole factor to evaluating camera profiles. This is very complicated color science here, and I don't understand it yet.

[Tim Lookingbill]

I'ld say if you want to test each profiling method for objective accuracy, pick colors that push the gamut capturing capability of the sensor itself as a way to rule this out when certain colors don't reproduce as intended.

Go to any office supply or Walmart and purchase this...

http://i.walmartimages.com/i/p/00/03/65/00/10/0003650010163_500X500.jpg

...take a shot of it and try to reproduce this vibrant aqua under any given light, using any profiling package or methodology even tweaking the color tables if necessary. If you don't get a spot on reproduction, use your profiling hardware to measure this aqua and plot its gamut. Compare this to DxO's gamut measurements of your camera model. If it shows it fits inside it then you have to question DxO's findings and methodology.

I have that Georgia Pacific package and I pretty much got the results shown in the link lighting it under fluorescent vs my camera's flash, applying a fluorescent DNG profile, dual & single table and Adobe Standard and couldn't get any closer. This is a VERY deep, rich and vibrant aqua. No where online can I find an accurate representation of this aqua.

I only have an sRGB-ish gamut display so those who have wide gamut may be able to reproduce this aqua and if not then you know that color's gamut is beyond any digital reproduction technology at this time and maybe even scientific instruments such as a spectro.

Here's my results...

[Vladimirovich]

Look, I have no ulterior motive in all this. I am just another natural light photographer chasing elusive perfection. If we realize that in all this camera profiling is a complete farce, I will be happy to swallow that. Right now I am honestly a bit confused as to how we are going to approach this testing.

but w/o an approach we shall not discuss... if it is a matter of your taste then there is nothing to discuss really... more over if you are a natural light photographer ask yourself hard - what do you not like in Adobe's profiles... matrices or LUTs really... you do not need any targets to get rid of LUTs or change them... why bother.

And for florescent lights the Adobe Standard profile seriously sucks.

you are a natural light photographer, are you not ?

Andrew, I am sure that with reference values from various color targets, we can evaluate the accuracy of the profiles by comparing with rendered values of photographed targets? But scene colors that are not on a target are going to be harder

suggestion was - build profile from one taget, measure the qualify from a different target - not the one that was used to build a profile !

[digitaldog]

suggestion was - build profile from one taget, measure the qualify from a different target - not the one that was used to build a profile !

It's a good suggestion and could be done or added to whatever testing process is worked out.

The initial idea I think was to see how ICC vs. DNG profiles work in terms of a single raw converter that can accept both. We don't have that option in Adobe raw converters. It is possible in Raw Developer and Brian the author give some tips on how to do this. One could shoot a scene raw then apply both types of profiles and post them side by side with the photographers personal preference. Then apply the same two profile types to other images and see how they do.

Someone could certainly capture say a Macbeth target, use both profiles, end up in say ProPhoto RGB and compare the reference (what those 24 patches should be) against the results and end up with a dE values. Then if camera metameric failure raises it's ugly head, we can ponder if using subjective subjects or as Tim illustrated, very difficult color originals is affected with one or both profile types equally.

[Hening Bettermann]

> It is possible in Raw Developer and Brian the author give some tips on how to do this.

?? In your quote in the Shooting Color Targets thread, I think he said Iridient Developer did NOT support DNG profiles other than embedded in DNG image files. So how would you embed them unless in ANOTHER raw converter that supports this? Or do you have new information?

[digitaldog]

> It is possible in Raw Developer and Brian the author give some tips on how to do this.

?? In your quote in the Shooting Color Targets thread, I think he said Iridient Developer did NOT support DNG profiles other than embedded in DNG image files. So how would you embed them unless in ANOTHER raw converter that supports this? Or do you have new information?

Correct (about only supporting embedded DNG image files). It appeared however one could test those profiles unless I'm having a brain fart and missing something:

I'd like to see good peer review on this subject! I'd like testing to be well defined. I'd like to see a number of savvy users (not the DP review crowed) to give the two processes a try then report their findings. Can this be done in your product, ideally even in demo mode? If so, are there specific steps one should follow to build both sets of profiles and then use them?

Demo mode will watermark images with red text which could throw off profiling tools... However, given a custom ICC profile for an image you could load that in the demo just fine for viewing.

DNG currently can only be of the original color matrix type and must be embedded into a DNG image file. No support for loading of standalone DCP format color profiles at this time in Iridient Developer at this time. So you need to profile the DNG and then embed the custom profile back into the DNG file. Again could load custom DNG profile data out of a DNG file for viewing with the demo, but I do not support export of DNG files for profiling.

for Iridient Developer to output an image for profiling:
1) Disable all color management by checking the checkbox on the Out pane of the settings window.
2) White balance the reference image if necessary
3) Adjust camera tone curve either to accurate linear gray scale or gamma 2.2 or sRGB or I've even done some work with LAB grayscale too (close to gamma 2.4). Some tools will be picky about the tone curve and will expect image data in roughly 2.2 or 1.8 gamma. Some like ArgyllCMS will work with just about anything you toss at them.
4) Export as TIFF ideally 16 bits/channel (critical for linear data) if the profiling tool supports it.

After profiling the created profile will generally be tied to the camera tone curve used above. Uncheck the disable color management option and choose the ICC profile in the Input Profile popup menu after copying to one of system ICC profile folders (/Library/ColorSyc/Profiles/ etc). You can modify the camera tone curve, but you'll lose accuracy just like you do if you modified an RGB tone curve to a carefully calibrated image from a scanner.

Most of my default camera profiles where I am not be hyper sensitive to getting accurate color I'll actually just go and modify the camera tone curve to normalize exposure and smoothly roll off highlights, etc even though the ICC profile was created under the assumption of perfectly linear data. As long as the curve is largely linear especially through the critical mid tones I think this is OK...

Or just skip ID and find another product that does both. Or wait on Brian to maybe allow this functionality which *could* happen.

[Hening Bettermann]

As far as I can gather, you must be referring to this passage:

>>"So you need to profile the DNG and then embed the custom profile back into the DNG file."

- this would have to be done in a converter other than ID, or is it me who is missing something?

then you could VIEW the result in ID:

>>Again could load custom DNG profile data out of a DNG file for viewing with the demo, but I do not support export of DNG files for profiling.
--
Raw Therapee has been reported to do both.

[Hening Bettermann]

@ samueljohnchia, original post
> Let's add to this list

Have you dropped the no-profile approach by ColorPerfect that you mentioned yourself? I find it gives very convincing results on scanned 3f-files.

BTW it strikes me that the monitor profiling software ColorEyes seems to follow the same philosophy, it just deals with grayscale and red, green and blue.

[Vladimirovich]

So how would you embed them unless in ANOTHER raw converter that supports this? Or do you have new information?

you open the file in ACR w/ the necessary camera profile and save it as DNG (non linear) for example... the profile shall be embedded then

[Hening Bettermann]

Yes of course! Thank you.

[Vladimirovich]

Raw Therapee has been reported to do both.

actually in addition to that we have quite active RT developers among forum members - at least two - Michael Ezra and Emil Martinec ... not sure if they worked on color profiles part there, but they might be able to comment.

[samueljohnchia]

Have you dropped the no-profile approach by ColorPerfect that you mentioned yourself? I find it gives very convincing results on scanned 3f-files.

BTW it strikes me that the monitor profiling software ColorEyes seems to follow the same philosophy, it just deals with grayscale and red, green and blue.

Hening, no I have not dropped ColorPerfect. I am trying to get up to speed with ColorPerfect, and have been corresponding with David to clarify certain image quality issues I noted. I have used it for digital raw images, not for scanner files. Generally speaking, it gives wonderful results at its default settings when compared to ACR's default settings. It also does not have a hidden baselineexposure offset and the color rendering is quite good considering that there is no typical "profile" of any sort, just balance for grays and let the colors fall where they may. However, I have gotten used to my custom camera profile and the way it renders color in ACR, and rightly I am biased. I tend to shy away from the default settings in ACR (too much contrast and saturation) and prefer to begin with a linear tonal representation of zeroed settings. I will comment more on ColorPerfect when I have a better grasp on how to use it and what it does. The current UI is so user-unfriendly that it is hindering my learning of its novel approach. I am also unsure if I would see improvement in the results if I custom gray calibrated my camera instead of using the generic calibration in ColorPerfect.

I appreciate the simplistic approach to all this. You always gain some and lose some in a profile. For display calibration, with low end displays, twisting it's native behaviour with a LUT profile may introduce artifacts, especially for low end displays. Question is, is a DNG or ICC camera profile ruining your "color integrity" - the main argument behind ColorPerfect's alternative approach?

you are a natural light photographer, are you not ?

I am open to learning about how camera profiles behave under different lighting situations. In an earlier post, I expressed interest in a copy studio setup with artificial lighting, despite my favor for natural light. In addition, I was interested to see if my new QPcard 203 really outperforms the CC under that illuminant, because that was what it is marketed to do. Untwisting the standard profile and comparing it alongside was a simple matter, and I was curious about how it performed. I think it would give us all a broader picture of what works when, if we look at a greater number of situations that just the ones we face.

Yes, I am also interested to learn how RT deals with DNG/ICC profiles, in particular the former and its dual matrices.

[Vladimirovich]

In an earlier post, I expressed interest in a copy studio setup with artificial lighting, despite my favor for natural light.

but then you do not need to intentionally use an artificial light w/ spectrum like sodium vapor or fluorescent light, do you ? why 'd you equip a studio w/ such light instead of a something w/ proper spectrum ? for some special effects ? but then you might be exactly looking for a bad color reproduction for that purpose 

[Vladimirovich]

It also does not have a hidden baselineexposure offset

you can manually set baseline exposure to zero in your .dcp profiles - so that is not a reason at all