GSoC CM collaboration

[Hal V. Engel]

On Sunday 02 March 2008 12:01:45 Maarten Maathuis wrote:
> On 3/2/08, Hal V. Engel <hvengel at astound.net> wrote:
> > On Sunday 02 March 2008 11:28:39 Maarten Maathuis wrote:
> >  > On 3/2/08, Kai-Uwe Behrmann <ku.b at gmx.de> wrote:
> >  > > Am 02.03.08, 19:56 +0100 schrieb Maarten Maathuis:
> >  > > > On 3/2/08, Kai-Uwe Behrmann <ku.b at gmx.de> wrote:
> >  > >  > > Am 02.03.08, 19:14 +0100 schrieb Maarten Maathuis:
> >  > >  > >  > What kind of color correction do you have in mind? (besides
> >  > >  > >  > the usual gamma adjustment)
> >  > >  > >
> >  > >  > > Many colour transformations are CLUT based. This is in case of
> >  > >  > > X a 3 dimensional table to interpolate from input to output.
> >  > >  > > Additional gamma and matix operations can play a role. But for
> >  > >  > > LCD's more and more CLUT's are used, as these devices are
> >  > >  > > sometimes very non linear other than CRT's.
> >  > > >
> >  > > > If all you want is control of the LUT's, then i suggest you look
> >  > > > at
> >  > > >
> >  > >  > randrproto. I'm not sure if it's perfect yet, but it should be
> >  > >  > reasonable.
> >  > >
> >  > > You talk about the graphics card gamma tables, while I talk about 3
> >  > >  dimensional tables. Usually the a 3*3*17 sized for monitors and
> >  > >  additional one dimensional curves.
> >  > >  A 3*3*17 cube needs some interpolation routine to get the final
> >  > >  output from an argitrary input. A nearest match would not suffice.
> >  >
> >  > Why 3d tables if i may ask?
> >
> > Because it is a 3D problem. For output devices like monitors color
> > management maps from some absolute color space such as CIELab or CIEXYZ
> > into the devices color space in a way that corrects all colors not just
> > those along the neutral axis.  The most you can do with the video card
> > LUT is to get the per channel gamma to be well behaved and the R=G=B axis
> > to be close to neutral. You can not get colors correct for R!=G!=B.  
> > This is a direct result of the 1D limitation of the video card LUTs.
> >
> >  An example, where this becomes very apparent is with the newer LED based
> > wide gamut monitors.  Even with a well calibrated video card LUT the
> > display colors where R!=G!=B, with out full color management, will be
> > much too saturated to the point of being garish.  These monitors are
> > starting to become fairly common so this will only become more of an
> > issue going forward.
>
> You're essentially assuming that the individual color channels are not
> linearly independent?

I am not sure what that means.  CRTs are fully additive devices and you can 
predict the actual color output of the device based on the transfer functions 
of the individual channels.  For many LCD devices with the possible exeption 
of some very expensive ones this is not true (IE. they are not fully 
additive).

The real point is that if you place two monitors side by side, even with 
calibrated video card LUTs and the same luminosity, displaying the same RGB 
values on the two devices can result in significantly different results 
unless the devices have very similar color characteristics.  For example if 
you were comparing a comsumer grade LCD display to a high end LED light LCD 
the colors will be dramatically different.  

This also means that if we are playing a video on these two devices and the 
video playback software is not color management aware (and I don't think any 
of these are at this time) then the video will have a totally different look 
on the two monitors.  Video playback software generally makes some 
assumptions about the color characterisicts of the display.  In the case of a 
a well calibrated CRT (including the LUT) this assumption would likely be 
fairly close to correct.  But for a wide gamut LCD this assumption is very 
far from correct.   

In addtion, all non-color management aware applcations (this is almost 
everything other than photo editing and publishing software on our desktops 
and Firefox 3!) make an implicit assumption that the display will have color 
characterisics fairly close to sRGB.  Again with many of the newer monitors 
this assumption is not even close to correct.

There is a fairly long thread on the OpenICC email list about this subject 
recently.  This thread was started by Chris Murphy on Feb 10 if you are 
interested in having a look at it.

>
> >  > >  The graphics card gamma tables are one dimensional, containing
> >  > > usual 3*256 values.
> >  >
> >  > True.
> >  >
> >  > >  kind regards
> >  > >  Kai-Uwe Behrmann
> >  > >  --
> >  > >  developing for colour management
> >  > >  www.behrmann.name + www.oyranos.org
> > >
> > > _______________________________________________
> > >
> >  > xorg mailing list
> >  > xorg at lists.freedesktop.org
> >  > http://lists.freedesktop.org/mailman/listinfo/xorg
> >
> >  _______________________________________________
> >  xorg mailing list
> >  xorg at lists.freedesktop.org
> >  http://lists.freedesktop.org/mailman/listinfo/xorg