Re: Possible bug in display profiles auto-generated from EDID: 'chad' tag converts from PCS to device

[Richard Hughes <hughsient gmail com>]

On 6 August 2011 20:07, Florian Höch
<lists+gnome-color-manager hoech org> wrote:
> "This tag, which must be invertible, converts an XYZ colour, measured at a
> device's specific illumination conditions, to an XYZ colour in the PCS
> illumination conditions after complete adaptation."
> So, the inverted matrix should do the opposite.

Agreed.

> This is not the case with the profile I looked at. When I multiply the
> inverted matrix with D50 XYZ numbers, which should yield the device
> illuminant (CCT around 6460K, I know it from looking at the EDID), I get a
> far too 'warm' result (CCT around 4408K). Not inverting the matrix yields
> the correct values.

Hmm, that sounds like a bug. I'm at the desktop summit at the moment,
but I can verify that myself in a week or so when I get back to my
development box.

> I'm not sure if this is a GCM or LCMS bug. I can provide the ICC profile in
> question (and a few 'reference' profiles from other profiling solutions
> which seem to have correct 'chad' tags) if needed.

I'm sure I'll have the same problem on my machine.

> The GCM version I used is from Fedora 15 (3.0.0-3.fc15). I skimmed the
> commit messages of the git repo to see if it's maybe already fixed upstream,
> but couldn't find anything.

Well, when creating the profile, GCM doesn't actually create the
chromatic adaption state itself, it relies on the cmsCreateRGBProfile
and the magic of lcms2. It's entirely possible that I'm doing
something wrong here, and an expert eye on the code would be a very
great thing (hint, hint) :-)

The latest code for 3.2 is here:
http://git.gnome.org/browse/gnome-settings-daemon/tree/plugins/color/gsd-color-manager.c#n539

Beers promised for bugs.

Richard.