[argyllcms] Re: Monitor calibration

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Thu, 23 Jul 2009 13:37:01 +1000

nome cognome wrote:
Spewctraview II is not a masterpiece, in fact if I run "dispcal -v -yl
-t6500 -g2.2 -qu calibrazione" (after the hardware calibration
performed with spectraview II) and I choose "2) White point (Color
temperature, R,G,B, Gain/Contrast)" I get DE  2.7 which is not so good
(I'm sure that Spectraview II performs hardware calibration, I used
X-Ryte Calibration (LUT) Tester to see if the video card LUT was
linear).
Target Br 133.11, x 0.3128, y 0.3291
Current Br 133.09, x 0.3074, y 0.3246  DE  2.7  R++ G+  B-

Some monitor specific calibration systems use custom instrument
calibration matrixes to achieve better accuracy from a colorimeter
that it's default generic calibration, so the above discrepancy
may actually reflect the inaccuracy of the generic instrument calibration
matrix that Argyll has to use, rather than any deficiency in Spectraview.

So I decided to perform also a software calibration (my video card is
an Ati Radeon HD 3870 which has a 10bit LUT). I typed "dispcal -v -yl
-t6500 -g2.2 -qu calibrazione" again and I pressed 7 to continue on to
calibration.
I did a little test to see if if I have obtained benefits by adding
Argyll software calibration:
1) Only hardware calibration + Argyll profiling
targen -v -d3 -f200 -A.5 -c"LCD3090WQXi 88100898YB.icc" monitor_140_m_1
LCD3090WQXi 88100898YB.icc is the profile created with Spectraview II.
dispwin -c
dispread -v -yl monitor_140_1
colprof -v -qm -ni -no -np -as monitor_140_1
targen -v -w -d3 -e 1 -g 32 -f 320 -r check_monitor_140_1
I created a new test chart to avoid the use of the same patches used
for profiling.
dispread -v -yl check_monitor_140_1
profcheck -v -k -w check_monitor_140_1.ti3 monitor_140_1.icm >
profcheck_monitor_140_1.log
Profile check complete, errors(CIEDE2000): max. = 1.926604, avg. =
0.308265, RMS = 0.380178
With only the hardware calibration (which has DE  2.7) the average error is 0.3.

2) Hardware calibration + Argyll software calibration + Argyll profiling
dispcal -v -yl -t6500 -g2.2 -qu calibrazione
targen -v -d3 -f200 -A.5 -c"LCD3090WQXi 88100898YB.icc" monitor_140_2
dispread -v -yl -kcalibrazione_140.cal monitor_140_2
colprof -v -D"monitor_140_2" -qm -ni -no -np -as monitor_140_2
targen -v -w -d3 -e 1 -g 32 -f 320 -r check_monitor_140_2
dispread -v -yl -kcalibrazione_140.cal check_monitor_140_2
profcheck -v -k -w check_monitor_140_2.ti3 monitor_140_2.icm >
profcheck_monitor_140_2.log
Profile check complete, errors(CIEDE2000): max. = 1.972252, avg. =
0.802361, RMS = 0.838653
With hardware calibration (which has DE  2.7) + Argyll software
calibration (Ultra quality) the average error is 0.8!

A couple of comments:

   200 points is a small number for profiling, although it
may be sufficient for a shaper/matrix profile. It's probably
too small a number for verification checks though. 1000 would
be better. I'm not sure why you are referring to the DE 2.7,
which is the hardware calibration white point delta E, since
the profiles will represent the behaviour of the device as
it is, and hence are not affected by any absolute white point
inaccuracy's.

The shaper or gamma/matrix profiles have a deficiency in
not perfectly fitting non-zero black device responses,
and so the shaper/matrix profile fit may be influenced by
the presence of hardware calibration. Without access to
the device in order to look at all the details, it's hard
to explain exactly why certain combination work as they do,
but generally any DE under 1 is reasonable good. If higher
fit accuracy was desired, I'd be using a LUT based profile
(maybe using XYZ PCS) with a larger number of test patches (1000+),
rather than a shaper/matrix profile.

Is this what should I expect? DE 2.7 is not a good result, so I
thought that I would have gained color accuracy calibrating with
Argyll before profiling.

Calibration mainly has benefits for non-color managed applications,
such as windowing systems etc., and in making sure that basic
parameters such as white point and brightness of decorative surround
elements do not throw off color perception when viewing images
using color managed applications. In the case of badly behaving
devices calibration may make for better fitting profiles (particularly
if matrix profiles are used), but if the display is already well
behaved, then using two calibrations (one in the display and one
on the display card) may add little advantage, or simply throw
more randomness into the situation.

Graeme Gill.

Other related posts: