Automatic colour calibration using Argyll CMS

[Misty]

All right, starting first off with the IT8 chart itself:

it8_noglare.jpg (146.27 KiB) Viewed 7207 times

This is, even at a glance, obviously much better looking. However - I came across some interesting issues with other photos in this set, which suggests to me that I need to refine how I'm producing profiles.

That profile was created using the colprof command:

Code: Select all

colprof -v -D"2013-05-23" -al -qm DSC_0225 -O 2013-05-23.icc

The important two settings are:

Code: Select all

-al - produces a cLUT profile, with a PCS (Profile Connection Space) of L*a*b*; Argyll recommends this as generally producing the best results
-qm - Default quality, also recommended

This is the same as how I did it earlier. This worked great for many images, but photos containing bright elements had very bad results. Let's compare to this post I made four years ago, with the same LD.

korpilot-al.jpg (429.01 KiB) Viewed 7207 times

Most of the image looks great, but the very bright yellow obi on the left was corrected to be overly bright, resulting in some really bad, ugly clipping. The colours weren't actually clipped in the original photo, so this is a problem with how the colour profile was generated.

I experimented with a few other profile types Argyll supports:

Code: Select all

-am - produces a profile with linear gamma, which is specifically listed as being useful for raw photos.
-ax - produces a profile with a PCS of XYZ, which it says "may have the advantage of better accuracy for additive type devices (displays, scanners, cameras etc.), may avoid clipping for displays with a colorant chromaticity that can't be encoded in L*a*b* PCS space, and may give a more accurate white point for input devices by avoiding clipping of values above the white point that can occur in L*a*b* based cLUT input profiles"
-aX - a variation on -ax. Argyll says that "If it is important for both the cLUT and matrix be accurate, use -aX, which will create shaper/matrix tags."

Here are the results of these:

-am:

korpilot-am.jpg (414.47 KiB) Viewed 7207 times

-ax:

korpilot-ax.jpg (414.02 KiB) Viewed 7207 times

-aX:

korpilot-acapsx.jpg (414.03 KiB) Viewed 7207 times

There's no discernable difference between the output of -ax and -aX.

All of these produced great results, and it's difficult to tell the difference after converting to sRGB. Even in a larger space like ProPhotoRGB, the differences are very minor, mainly that:

• The colour tone of Madoka's pink shirt is slightly different (in -am it's a little deeper, whereas it's a little bit brighter in -ax/-aX). In -am it's about 247/184/207, whereas in -ax it's 255, 195, 222. This indicates that that colour region is being clipped in -ax.
• The purple-red background is darker in -qm, where it's about 197/88/102, than in -ax, where it's 206/84/105.
• The whites and lighter tone of the obi are brigher in -am than in -ax.
• The darker skin tones are darker in -ax than in -am.

(All colour values given are after converting to ProPhotoRGB.)

Overall the colours in -qm appear to be more accurate to the real LD when I hold it in front of me. However, there's serious clipping in all of them - perhaps ProPhotoRGB simply isn't able to represent all of these colours? The original image, before converting, has no clipping beyond some light reflections on the very left edge of the obi.

[Misty]

Ugh, one would have hoped that I would have remembered by this point in my life about different rendering intents when converting colour spaces, no? Absolute colorimetric intent, in this circumstance, prevents most clipping when converting to ProPhoto RGB, and results in a remarkably accurate colour reproduction. I could, I presume, improve on the clipping further by selecting a wider-gamut profile - most likely ProPhoto RGB can't cover everything that the camera is capturing even with absolute colorimetric.

Here's an sRGB sample which, clipping aside, very faithfully reproduces the colours of a) the original, b) the absolute colorimetric conversion into ProPhoto RGB.

korpilot-absolutecolorimetric.jpg (396.82 KiB) Viewed 7200 times

I'll do further tests with other images and provide more samples. I'm hopeful that I've come across The Secret for high-accuracy completely automated colour calibration.

[abmartin]

Brava!

It's astonishing what a difference eliminating ambient lighting makes! Seeing that photo of the calibration card is the perfect example of why light shrouds are necessary. (I miss mine, and should really come up with a new system)

Thanks for all those great examples! (And there's just something strangely awesome about Japanese Laserdiscs!) The differences between the different approaches is astonishing (and I'm color blind!). I look forward to seeing more as you develop your approaches. I've already learned a lot. I didn't know about ProPhotoRGB before, for example. I believe that you are using libre tools, making this a real game changer for archives. I've seen a lot of archive photos with color calibration charts included that aren't calibrated accuractely and those were by groups that have extremely expensive equipment and software setups.

I am looking at some of the inexpensive color charts so I can start playing a bit too. This absolutely destroys the white balancing and brightness adjustments I am doing with gray cards. Adding options like this to my preprocess script is extremely exciting.

[Misty]

Wolf Faust sells really affordable IT8 charts that will work great with Argyll! He even has a non-glossy one.

Yep, all 100% libre! The primary tools so far are Argyll (AGPL3) and dcraw (GPL2); additional tools in the script for this will also be 100% libre too. I think it'll really be useful to have a system that requires no expensive software, affordable colour targets, etc.