[argyllcms] Re: Gamma wrong for calibration curves?

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sat, 02 Dec 2006 19:25:39 +1100

william.wood@xxxxxxx wrote:

The source profile in my tests is sRGB - gamma 2.2 (or close) isn't it?
> So sRGB converting to Monitor profile (perceptual) should not involve a gamma
change, yet images noticeably darken when I apply the profile conversion.

Well it depends on all the other details (type of link etc.), but it's
all going to be completely independent of the calibration. If you don't
think what's happening there is right, then you need to investigate the
profile accuracy, and what happens to the colors as they are transformed
though the linking process, to decide if there is something wrong with
that aspect.

I'm getting the same results with -k0 and -k1; resulting callibration curve
gamma is 2.05, not 2.2.

OK, so your display has a black point with a good match in
color to the white point.

>  Below I've listed the output from dispcal -E for
.cal files with k=0 and k=1.  The gamma offset is the same (~0.08) in both
cases.  I don't understand why the gamma offset is low yet the gamma seems
so far off.

I:\Util\Argyll\profiles>..\bin\dispwin -d2 t42pk1.cal
I:\Util\Argyll\profiles>..\bin\dispcal -v -d2 -yl -E
Maximum neutral error (@ 0.095629) = 1.150478 deltaE
Average neutral error = 0.447931 deltaE

Right, a nice tight verification. The target curve may or may not
be what you want or expect, but the calibration has made the display
conform the target curve.

Here's the output from dispcal -E on a .cal file with a k of 0:
I:\Util\Argyll\profiles>..\bin\dispwin -d2 t42pk0.cal
I:\Util\Argyll\profiles>..\bin\dispcal -v -d2 -yl -E

You need to specify -k0 for dispcal when verifying, otherwise
it will create a k = 1 target curve to verify against and
will get ...

Maximum neutral error (@ 0.112249) = 6.966823 deltaE
Average neutral error = 3.416810 deltaE

a poor verification fit.

The discrepancy in gamma values you note is fully explained by
the fact the calibration target curve is a truncated ideal
gamma curve. Rather than subtract the black level (assuming view
black adaptation, which is a poor assumption), the target curve
is an ideal gamma curve, truncated at the point the display
can't go any blacker. If the gamma number created using -r
was estimated on the assumption that the curve in question is
a truncated ideal gamma curve, then the agreement with the
calibration target is within the tolerance of the calibration
(something I've confirmed by making just those alterations).

Now we can have a whole discussion about whether a truncated
ideal calibration curve is a good thing or not, but at the end
of the day anyone interested in color will be using profiles
on top of the calibration, and this renders such details less
important, as long as the calibrated display has controllable

Graeme Gill.

Other related posts: