[argyllcms] Re: Profile input white not mapping to output white

  • From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 26 Nov 2012 22:05:41 -0700

An update to the follow-up.

After re-doing the shoot (but with insignificant color-related changes), I've 
spent the past several hours testing all sorts of permutations of profile 
generation, using both my chart and a ColorChecker Passport (using all 50 
patches, not just the classic 24), the 1.4.0 version of colprof and the one 
Graeme just posted a link to, adding and removing various patches, and various 
options to colprof. I made a spreadsheet of a bunch of relevant numerical data, 
and, in Photoshop, assigned each profile to the original TIFF and visually 
compared the results.

There is one and only one factor that is consistent in improving profile 
quality, and it's a dramatic one. There is no question, period, full stop, but 
that adding the following to a .ti3 vastly improves the resulting profile:

WHT 96.4220 100.0 82.5210 100.0 100.0 100.0 0.0 0.0 0.0

A significant improvement, but not quite as good, was instead adding a line to 
the .ti3 of the measurement of the ambient light (which happened to be: 
X=100.183453 Y=104.100151 Z=90.705105 / L=101.564204 a=-0.319798 b=-3.733409, a 
value curiously close to D50 white -- coincidence?).

Removing the black trap made no visible difference in the results, and only 
made insignificant changes to the numbers.

Adding an all-zero line to the .ti3 moved the black point to the neutral axis, 
but there wasn't any significant change (though there might be a hint of better 

All of the preceding applies to LUT profiles.

The ColorChecker did not produce an acceptable profile using a LUT. With the 
ColorChecker and a matrix / shaper profile, there was no visible difference 
with and without the fake D50 white sample. At a first glance, the results with 
the ColorChecker are not bad. However, the ColorChecker profile can't properly 
render the black trap in my chart no matter what I tried, and there are a 
number of color shifts throughout the range (though luminance is fairly 
consistent except for the shadows). Still, it might prove acceptable with a bit 
of manual tweaking under certain non-critical circumstances.

Graeme, I think the new extrapolation code needs a bit of work. It didn't 
always render input white as output white, and there were times when it seemed 
like there might have been some inversion going on in the highlights. It did do 
quite swimmingly when fed a profile with an added D50 patch. Oh -- and you 
hinted at this, but there doesn't seem to be any difference between profiles 
with and without -u.

I'll close with a few attachments: a JPEG with the same values as the original 
TIFF, tagged with the winning profile (added D50 and black patches, no options 
to colprof); synthetic versions of my chart and the ColorChecker Passport for 
visual reference; the two .ti3s (with the D50 white and black lines added at 
the bottom); and a LibreOffice spreadsheet with my results. Of course, I'm more 
than happy to supply any other files I have should anybody think them relevant.



P.S. I can't see any difference between either chart in the synthetic or 
photographed version when doing a blink comparison, which is exactly the end 
result I desired. b&

P.P.S. I just did one more test, merging the ColorChecker Passport .ti3 into 
the .ti3 from my chart, with both D50 white and black patches. Visually, there 
was no change and the numbers were slightly worse. The yellow at L22 isn't 
quite as saturated, for example, but I think that's outside my monitor's gamut 
and I can't see the difference. It could also be Photoshop to blame. b&




Attachment: TrumpetChart!.ti3.D50white0black
Description: Binary data

Attachment: ColorChecker Passport.ti3.D50white0black
Description: Binary data

Attachment: Profile comparisons.ods
Description: application/vnd.oasis.opendocument.spreadsheet

Other related posts: