Camera Profiling - DNG, ICC and alternative methods

[samueljohnchia]

This is a continuation of the discussion from Hening's Shooting Color Targets thread, as it digressed into different approaches to camera profiling. I would first like to say that I don't think that there is anything out rightly wrong with DNG/ICC profiles or one method of working compared with another. Each will have its own limitations, and it would be unfair to demand performance outside these limitations. It is surprising that until now nobody has done any comparative testing on various approaches to camera profiling and subsequent raw processing for color.

It would probably be necessary to unravel how DNG/ICC profiles are made, and how raw converters utilize these profiles and make the necessary decisions to transform or extrapolate color. Some portion of that would likely be propriety and would require reverse engineering or inside knowledge, for us to be able to comment better on each approach without being uncertain about where any fault may lie.

I shall begin a list of interesting areas for potential testing:

DNG Profiles
ColorChecker + Adobe DNG Profile Editor
ColorChecker + X-rite ColorChecker Passport Software
QPcard 203 + QPcalibration
(Datacolor's solution is absolute bosh, let's avoid that please.)

ICC Profiles
ColorChecker + ArgyllCMS
ColorChecker SG + ArgyllCMS
QPcard 203 + ArgyllCMS
QPcard 203 + QPcalibration (with additional ICC plugin)

Raw Convertors
Irident Developer
Raw Therapee
Possibly just ACR/Lightroom to compare DNG profiles and "neutral state" rendering of the 800 pound gorilla
Capture One? Lots of folks claim that they get "better color"


Let's add to this list, and also discuss how we are going to design the testing for all of it.

[Vladimirovich]

Adobe does not make its own camera profiles using those targets and Adobe PE... "Cura te ipsum" (c)

[samueljohnchia]

Adobe does not make its own camera profiles using those targets and Adobe PE... "Cura te ipsum" (c)

Could you enlighten us on how Adobe makes their profiles?

[Schewe]

Could you enlighten us on how Adobe makes their profiles?

Actually, Eric Chan and Thomas Knoll make the DNG profiles using some slightly more exotic hardware and software. Exactly how they do that and what they use has not been disclosed...you would have to ask them.

[digitaldog]

While it would be fascinating to hear how Adobe builds their profiles, for these tests it hardly matters since the rest of us can't do so. It is still interesting to compare a user created profile with the tools outlined here to the canned profiles.

(Datacolor's solution is absolute bosh, let's avoid that please.)

Indeed!

[Stephen G]

This is all very close to my work so I'll be following this thread. I profile for art reproduction. Even though I don't have the technical background to contribute at a high level I'll share my experiences of what has produced good results for me.

Adding a few items to the lists:

Raw converters
DXO Optics Pro. DXO can produce either linear or gamma corrected ("realistic color rendering") output files specifically intended for camera profiling. I've gotten very good results with profiles built from the linear output.
PhotoNinja. Has built in profiling for 'camera sensor' and 'light source'. Never tested it, know nothing about it.

ICC profiles:
HCT + ArgyllCMS

notes:
RAWTherapee does support both DNG and ICC profiles and has options to blend the two.
CoCa is a nice camera profiling GUI for ArgyllCMS. Lots of options, easy to use, can use wide variety of targets.

[samueljohnchia]

This is all very close to my work so I'll be following this thread. I profile for art reproduction. Even though I don't have the technical background to contribute at a high level I'll share my experiences of what has produced good results for me.

Thank you for your contribution Stephen. This is wonderful as you already have a setup for making reproductions and can therefore test for a studio environment, especially one that is setup for reproduction work (you must have been more critical about your lighting). I am a natural light photographer and own no studio lighting, so I can only test for daylight.

Do you happen to have a ColorChecker SG besides your HCT chart?

[Stephen G]

Happy to join in - helps me refine what I'm doing.

I don't have a Colorchecker SG, just the HCT 5x7 reflective, CC24 mini and a QP203. Wish I did have a CCSG as the scanner profile I've built from it (I rented one) is superb.

My lights are not wonderful, but they do the job. I've got two banks of three halogen spots that I move around on lightstands. Hoping to get some SoLux bulbs soon, but getting them here is expensive. I eliminate all other light in the room, however, and flat-field with Equalight before profiling.

[Vladimirovich]

Could you enlighten us on how Adobe makes their profiles?

http://en.wikipedia.org/wiki/Monochromator in 5nm/10nm increments across the spectrum

as for software they did not tell you, but if you research postings from Eric Chan (try http://forums.adobe.com/community/cameraraw ) you will see that he specifically referencing that Adobe PE can't build profiles that Adobe builds to bundle along w/ ACR/LR

and when somebody comes out swinging that he can do better for a regular light than standard profile I can suggest to leave in both profiles only matrices, remove LUTs (down w/ twists!), put linear tone curve (everything can be done w/ Sandy Mc dcptool and simple text editor) and think again which one is actually better... in most regular light cases you just need a proper WB instead of placebo in the form of $100 consumer level unmeasured target where in most cases you can't even illuminate it precisely in the field...

[digitaldog]

DXO Optics Pro. DXO can produce either linear or gamma corrected ("realistic color rendering") output files specifically intended for camera profiling.

"realistic color rendering"? You mean your linear gamma images look very dark and flat? That's because there is no correct profile assigned to that data. IOW, if you have a profile to define linear gamma data, it looks just like gamma corrected images.

[samueljohnchia]

and when somebody comes out swinging that he can do better for a regular light than standard profile I can suggest to leave in both profiles only matrices, remove LUTs (down w/ twists!), put linear tone curve (everything can be done w/ Sandy Mc dcptool and simple text editor) and think again which one is actually better... in most regular light cases you just need a proper WB instead of placebo in the form of $100 consumer level unmeasured target where in most cases you can't even illuminate it precisely in the field...

Unfortunately for me and my camera that I use that is not the case. Yes, I also tried untwisting profiles too. Andrew has also commented elsewhere in this forum that he has produced custom camera profiles that were better than the generic Adobe profiles.

Interestingly Ernst commented in my thread on QP that he would go with the camera manufacturer's "estimations of sensor responses". ColorPerfect is also based on a similar philosophy, in the sense that common folk like me cannot possibly render color better with these so-called profiling tools. ColorPerfect takes it further and claims that a gray balancing alone is sufficient - and avoids ICC and DNG profilles entirely, claiming they are a farce and destroy color integrity.

This is why we need testing. I want to see the proof in all this. Who is right? Is what I am seeing all an illusion?

[digitaldog]

Andrew has also commented elsewhere in this forum that he has produced custom camera profiles that were better than the generic Adobe profiles.

I'm not the only one. I've yet to see a custom DNG Profile that I didn't prefer over the generic profile. I don't think this has much to do with the different techniques we and Adobe use. IF, at least in my case, Canon's sensors were really consistent and behaved like the Canon sample Adobe used to build their profiles, their canned profile might very well be better than my DNG profile. I suspect it's why Adobe has provided us the tools to make our own DNG profiles.

You can have the most expensive Spectrophotometer, use the largest patch sample and best profile software to build a printer profile. If that ends up as a canned profile for another printer, it can work really well or less well then building a custom profile on your specific printer with a lesser quality Spectrophotometer or number of patches.

What Adobe uses to build canned DNG profiles would be interesting but not pertinent to his discussion which is looking at the differences in making custom profiles, DNG or ICC based.

[samueljohnchia]

Interesting to note, the dcp profiles by Adobe and the custom profile with the DNG PE - both have two camera and forward matrices. The matrix values are identical, at least for my camera model.

Adobe profile HueSatDelta LUT has 36 hueDivisions, 8 satDivisions and 16 valDivisions.

DNG PE custom profile has 90 hueDivisions, 25 satDivisions and 1 valDivisions.

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.

[Vladimirovich]

Andrew has also commented elsewhere in this forum that he has produced custom camera profiles that were better than the generic Adobe profiles.

he then can illustrate w/ a set of dEs otherwise it is his taste and taste is subjective, that's it (= placebo)

[Vladimirovich]

Interesting to note, the dcp profiles by Adobe and the custom profile with the DNG PE - both have two camera and forward matrices. The matrix values are identical, at least for my camera model.

that is because you use Adobe PE to tune the base (standard profile), that's it... speaking about placebos.

[digitaldog]

he then can illustrate w/ a set of dEs otherwise it is his taste and taste is subjective, that's it (= placebo)

It's totally subjective! What dE values would I be comparing and how useful would the lower one be if I didn't like the color?

[Vladimirovich]

Adobe profile HueSatDelta LUT has 36 hueDivisions, 8 satDivisions and 16 valDivisions.

DNG PE custom profile has 90 hueDivisions, 25 satDivisions and 1 valDivisions.

they call it 2.5D vs 3D profiles

[Vladimirovich]

It's totally subjective!

that is exactly the point - you can't argue w/ anything subjective... as soon as it is a matter of taste please by all means consider your profile better.

[samueljohnchia]

he then can illustrate w/ a set of dEs otherwise it is his taste and taste is subjective, that's it (= placebo)

I am sure he can. But a couple of color patches ain't going to cut it for you and maybe me and others too. What about difficult to reproduce colors? I'm not sure if we can spectral measure scene colors for our testing too.

A short take by Eric on optimizing the color rending of profiles: http://www.luminous-landscape.com/forum/index.php?topic=59688.msg482292#msg482292

[digitaldog]

that is exactly the point - you can't argue w/ anything subjective... as soon as it is a matter of taste please by all means consider your profile better.

Yes, as we find in a huge part of photography and image creation. I don't think the audience here has much interest in scientific capture of data, they (and I) want to make pleasing images. IF we gather a group to test a single raw converter that can utilize a DNG and ICC camera profile, their preferences will be subjective. Unless one or the other process produces what the image creator feels is an awful image (which I have seen with ICC camera profiles). Splitting hairs with dE values don’t belong here.