[argyllcms] Re: Camera matrix profile, adding ti3 perfect white data set

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Wed, 03 Feb 2010 14:23:22 +1100

Gerhard Fuernkranz wrote:

Hello Gerhard,


well, the reference file specifies the actual colors. For instance, the
GSxx patches on IT8 targets from Wolf Faust seem to be close to D50
(i.e. illuminant) chromaticity, but Dmin is unprinted paper and Dmax is
the darkest color that can be printed on the medium, so the chromaticity
of these patches can deviate a bit more from D50.


Sure, but it comes down to intent. If the test chart is representative
of the medium being scanned, then (doing things the ICC way), the medium
should map to D50 in relative colorimetric, and that's what someone
putting documents on a scanner probably wants. They don't want to
see the paper in the scans, just what's printed on it, so that they
can then reproduce it on some other medium (display, printer)
without the original paper showing up in it.

[It's an unfortunate consequence of the ICC's choosing to store
 the profile data as colorimetric that this also implies clipping
 any values above the white point, so that using such a standard
 profile in absolute colorimetric mode doesn't cover as bigger
 range as it might, although any extra ranges will be extrapolation
 anyway.]


But the actual issue is not really the color of the patches, but a
different one: Does it really make sense to place the media white point
of camera profiles at Dmin of the target? IMO it does not. We certainly
need a target in order to establish a RGB to XYZ mapping, but we don't
need the target for anything else. The target is not the "medium" we


Sure. But in this situation absolute colorimetric can be used,
so that the profiles choice of white point then has no influence.


If the camera already returns white-balanced raw RGB numbers and we
simply want to trust the camera's white balancing (regardless whether it
is correct or not), then the XYZ color corresponding to
RGB=[255,255,255] is IMO a suitable media white point. A relcol
conversion to a working space will map those colors to neutral which are
considered by the camera's white balancing as neutral. And no RGB
numbers will be clipped.


I'm not so sure that color profiles are useful when the camera
is doing white balancing. Even if one assumes that the camera
is simply doing a Bradford transform in XYZ space, I haven't
figured out then whether even something as simple as a calibration
matrix is applicable. And since the very assumption being made
is that the camera isn't properly calibrated, it can't possibly
be doing a white balance using a Bradford transform in true XYZ space :-)

If it's using some other "creative" white balance that "looks good"
in the resulting RGB pictures, all bets are definitely off.
I wouldn't image a calibration matrix made at one automatic
white balance point would necessarily be of much help when used after
a different mystery white point balance, and certainly
wouldn't be colorimetically accurate. (But those with more
practical experience can try and convince me otherwise, although
some theory to explain it would be good.)


The same media white point, located at RGB=[255,255,255], is IMO still
appropriate if the raw RGB numbers from the camera are not yet white
balanced at all so that we need to white balance each image anyway
individually after conversion from camera RGB to XYZ or to a working space.


If this is really useful, can't it simply be achieved by
setting the cameras white balance and exposure such that the white
test patch has a value of 255,255,255 ? (Or is that difficult
to do in practice ?)


If the raw RGB numbers are almost (but not exactly) white balanced by
the camera and we want the profile to fix the residual white-balancing
error, then rather a media white point with illuminant chromaticity is
what we want. Again the white point's luminance should be scaled so that
either min(R_wtpt,G_wtpt,B_wtpt)=255 or max(R_wtpt,G_wtpt,B_wtpt)=255
applies, in order that we still can utilize either the full RGB range
w/o clipping, or almost the full RGB range with slight clipping [the 2nd
variant with slight clipping has advantages too, as it avoids "chromatic
highlights" after conversion to a working space].


My assumption has been that using ICC profiles to do white balancing
for camera images (relative colorimetric) is not a practical proposition,
since in anything other than a fixed, studio situation the white
balance will be different for each shot, and therefore needs to be
done in a way that is separate to color management.

cheers,

Graeme Gill.

Other related posts:

  1. Home
  2. argyllcms
  3. Archive
  4. 02-2010