Accuracy of color rendition in converters - lightly quantitative

[Mark S Abeln]

My question was far more specific, but perhaps in that sense also misguided.

Let's say I do a DCP profile in adobe. I then apply the DCP profile to the color checker.

How do I know it worked properly?

Other than "looks OK by eye". Is there a practical, end-user technique that will tell me that, generally speaking, all the software and raw development steps in the middle more or less ended up with the "right" result?

The "right" result will be where each of the ColorChecker patches as photographed by the camera exactly match the actual values on the target itself, and not only that, but all other colors will match as well.

What you'll find in reality is that some patches in your photograph, after calibration, will match the target quite closely and others won't. This goes back to the 'figure of merit' problem I mentioned earlier. If you want a single overall value to measure calibration 'goodness' understand that this is not something that Color Science can deliver. Is having most colors completely accurate—with one way off—better than having all colors just OK with none really good? Or do you want good skin colors? Or do you want accurate colors that are not represented on the test chart?

But the actual target values have to measured under controlled conditions: the published values may not actually match your target, due to process changes, sample variation, and target deterioration.

Also, understand that color calibration typically attempts to provide colors as if they were photographed under standard conditions, for example 6500K color temperature in very bright lighting. But what if you want colors that will match how you saw them when you photographed the target? Good luck with the last one! It's possible to calibrate cameras that way, but I know of no one who does that.

That goes to the jist of it -- I'm fairly happy with the results, applied to real world photos.

I probably should stop there, but I didn't.

Hehehe, I went down the same road myself!

I also compared the same real world photos with other raw conversion processes. And individually, separately, I'm fairly happy with those results.

That's a great start.

My conclusion -- my eye is flexible, and easy to please. A.k.a undiscerning. Compare it to, say, a piano -- I could not tune one, and can barely play, but can definitely tell you when one is off. It kind of "hurts" to listen to. And like photos, piano tuning is not a simple, decided science -- you do not just follow the half ratios in each octave. There's art to it, and like color art I do not understand. 10 pianos could sound quite differently but all be "in tune" and sound ok. Maybe I prefer one to the other, but they do not have that almost "hurt" quality that a really out of tune one does.

Yet lots of people could listen to a very badly out of tune piano and enjoy it equally, and not hear the difference.

I'm not sure that is entirely true. Think of the old clanky piano sounds that you hear in old Westerns. It's distinctive, but not necessarily 'good'.

Others can see when colors are wrong, and I think they see it better than I. (Note I have been tested and I have no color blindness at all, just like people who might not know a piano is out of tune could test perfectly in all ranges for hearing).

So my question still goes to -- given I do not trust my eyes to be as discerning as I might like, are there techniques I can use to tell if the workflow I use is producing "right" results, so those with more discerning eyes will not go "yuck".

I think that you are 90% there with just the basic calibration you've done. Now there will be extremely discerning people who will want some key colors exact, and the only thing you can do in that situation is to force the colors to be correct. Understand that this may not work 100% perfectly, and there is indeed a great advantage to using high quality, low-noise gear with lots of lighting and great technique in order to do these corrections well. Fashion photographers do this sort of thing routinely.

Here maybe what I should be asking is what word goes where I used "right" that can be measured?

Or is there just no practical answer to that -- art remains art (at least within the means of an average hobby photographer)?

A good calibration should make the colors at least 'rightish'. For sure, you can expect cameras to be already fairly well calibrated under daylight, flash, and incandescent, but it is under gas discharge and LED lamps where you really run into problems.

A tool I have works much better for me than a tool I don't.

Hehe. Myself, I just learned to enjoy the results of the Adobe profiles (except when I was forced not to).

Seriously -- the answer to my question may be somewhere between "it's not a meaningful question" and "you do not have the wherewithal to answer it". That's OK too.

Yes, I think so.

I'll look, though that has been a very pleasant side effect of these DCP profiles. My D800 and D4 long ago had very different colors run through the same raw process in Adobe. After a DCP profile applied, largely the same (the AWB was also quite different but that's a different conversation). I've been doing it ever since, and colors, especially under jagged spectra lights, tend to be very similar between cameras.

I think that is the main purpose of profiling: making any camera give similar colors to any other.

Anyway... thanks for trying to educate the ignorant; I'm trying (in either definition of the usage).

Good luck! It is fascinating stuff, and there are true experts on this forum (not me) who really know how it all works.

 

[hjulenissen]

2. Adjust color rendering to (face tracked) head size ?

Hmmm not sure what you mean here.

I think that color appearance models depends on the size of the color patch? If so, it seems to make sense doing «something»(tm) to the color rendering depending on if the display (or a feature of the scene) covers 1 degree or 10 degrees of the viewers field of view?

-h

 

[MrBrightSide]

Quick note on evaluating color by eye. You only have a few seconds to judge the colors on screen before your brain starts adapting them to its own ideas of what they should be so your very first impression will be the most accurate. You can recalibrate your eye by getting away from the computer for a few minutes.

From a practical standpoint, for 95 percent of what I do, I get accurate results with two profiles, a daylight profile created using the sun at noon and a tungsten profile created using regular 100 watt tungsten-filament light bulbs. For the 5 percent that I know will be trouble, that's when I pull out the ColorChecker and create a new profile .

Here's an article by a guy who explains how he uses camera profiles to solve a very specific problem. Interestingly he prefers Adobe's DNG Profile Editor to X-Rite's.

So I think it is safe to say that when you're using a ColorChecker and the latest profiling software you are getting the best possible results for the largest number of clients. Is it perfect? Absolutely not and that's where judgement and your eye get their workout.

Getting back to the OP's question If you want to compare the values you're getting from your cameras to the values that X-rite uses for the patches, here they are. This is the 2010 version from X-rite, there may be a newer edition but I have not seen it.

And disregard the colors you see in these two charts. They're screen shots from X-rite PDFs. Only the numbers are real.

To me, the least understandable part of it is that they report black as 52/52/52 which is higher even than in 4-color printing.

And here's the disclaimer from the PDF containing the chart:

"The data in this file is reported in CIE L* a* b* data for illuminant D50 and 2 degree observer and is being supplied by X-Rite, Inc. This information can be used for personal and educational but not for commercial purposes without license from X-rite, Inc."

OK, but I'm feeling dense.... HOW do I compare?

For example, I took the a photo of the ColorChecker with an A7Riv under LED lights at a sports arena. I used that for a profile, then applied the profile to the shot, exported from Lightroom to Photoshop. In Photoshop I converted to sRGB.

I went to the Neutral 6.5 which calls for 160/160/160. I played with exposure and a bit of curve to get it near that (160,164,166).

The I looked up blue (13) which is (56,61,150). What I got was (0,41,191) so I'm clearly on a completely wrong track.

I saw the other posting about Patchtool, but am hesitant to buy something I don't understand without knowing if it will help me.

Here's the real goal -- I build Adobe DCP profiles in each arena or stadium under their artificial light, usually LED or Mercury Vapor. This seems to true up the colors by my eye and memory. It also seems to make different camera bodies match each other more closely.

When doing an outdoor arena I might shoot in daylight, I do a daylight + night dual illuminant profile. When doing an interior arena I just do the one color.

This seems to work well, mostly, but a new camera and comparing the results with other tools, e.g. their native application, Capture One (which is somehow affiliated with Sony I think) and Lightroom, they are quite different out of the box -- not just different in contrast and saturation and such which I expect, but in actual colors.

THe simple path is to pick the one I like the best.

Apparently the really complicated path is to ask how to tell which one is most correct.

Linwood