[argyllcms] Re: Am I stupid or is it something else... my color management sucks.

  • From: Magnus Berg <magnus@xxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sun, 11 Apr 2010 16:15:46 +0200

Thank you for your answer Graeme, and thanks for Argyll.

But I just found a quote from NEC that said:
"The Datacolor Spyder2 color sensor has been tested with the the NEC LCD2690WUXi, LCD3090WQXi and LCD2180WG-LED displays and found to cause inaccurate measurements when measuring the color primaries. This is due to the wide color gamut aspect of these displays. This may result in an inaccurate calibration and ICC Profiles to be generated. Using this color sensor is not recommended with these displays."

Still I learned a lot of your answer Graeme. My technical English is not that good and you gave me answer to a few technical questions I had.

Case closed!


2010-04-11 15:44, Graeme Gill skrev:
Magnus Berg wrote:
Hi, I'm new to the list and got a big problem.

I bought me a used Spyder 2. Realized that my old CTR couldn't produce enough light (90 cd/m^2 not 140 cd/m^2 that sRGB require). Digged a deep hole in my wallet and bought a quite expensive NEC 2690WUXI2. Now my color managed application show a green cast after calibration and proofing. It's not fun at all. :-(

If your instrument was giving good results on your previous display,
then it sounds like the problem might be that the Spyder 2 and the NEC
aren't well matched (but read on for other possibilities).

If this were the case, then a different colorimeter (i1 Display 2 or
perhaps a Spyder 3) may give a better result, but there are no guarantees
unless the colorimeter has been setup for the spectral characteristics of
that particular monitor. It would be advisable to test any new instrument
out on this display before committing money.

The only type of instrument that should give reliably better results
if this is the problem is a (more expensive) spectrometer type instrument
such as the ColorMunki.

This is how I do the calibration and proofing:

I use dispcalCUI. After that I had use the screen's OSD to set the 'White point', 'White level' and 'Black point' i got this nice values then I do a 'Check all':

Looks pretty reasonable.

Then dispcalGUI send this command to dispcal:
dispcal -v -d1 -c1 -yl -P0.500317863954,0.485345838218,1.0 -qh -t6500 -b140 -G2.2 -f0 -k0 -A4 -B0.0001 colprof -v -qh -al

I wouldn't use "-G2.2", I'd use "-g2.2" as it's more robust. A technical
gamma ("-G") tends to give results that are more sensitive to the black level.

Before, then I only did a -qm calibration, I got strong red casts on my calibrated screen, and the profiled applications gave strong green casts. The strange thing then was that the red cast only occurred on color 254,254,254 to around 249,249,249 but tones darker than that looked neutral grey with my eyes. After that I chose -qh this color tones and others looks neutral, with my eyes.

Sounds like the channel response is rather poorly behaved in that region.
Does the display have any setup pre-sets that give it better behaviour ?
(It is an LCD after all, so the behaviour is rather synthetic, and under
control of the electronics.)

After that I load the calibration profile done after the command above I got this then I run dispcal -r -yl:

magnus@debian:~$ dispcal -r -yl
XRandR 1.2 is faulty - falling back to older extensions
Place instrument on test window.
Hit Esc or Q to give up, any other key to continue:
Current calibration response:
Black level = 0.60 cd/m^2
White level = 135.50 cd/m^2
Aprox. gamma = 1.92
Contrast ratio = 226:1
White chromaticity coordinates 0.3099, 0.3266
White    Correlated Color Temperature = 6676K, DE 2K to locus =  4.9
White Correlated Daylight Temperature = 6676K, DE 2K to locus =  0.3
White        Visual Color Temperature = 6490K, DE 2K to locus =  4.7
White     Visual Daylight Temperature = 6664K, DE 2K to locus =  0.3
The instrument can be removed from the screen.

I understand nothing. How can I have gamma 1.92 and the above temperature then I chose to calibrate to 2.2 and 6500?

You chose a technical gamma of 2.2, and got an effective gamma of 1.92.
That's how it works, since the technical gamma gets offset by the
black point. See <http://www.argyllcms.com/doc/gamma.html>.

The white point seems relatively close (0.3 DE from the daylight locus).
There is no straightforward relationship between degrees K and delta E,
so a delta of 176 degrees may be quite acceptable. It's perfectly possible that the display has drifted between calibration and verification (backlights do that.) You (of course) need to make sure that none of the "features" are getting in the way (such as their "AmbiBright - ambient light sensor" which
presumably alters the display in response to ambient light changes).

And why is there a green cast on the pictures viewed in Color managed applications?

It depends on how accurate the profile is.

Did I understand things right:
I should load the profile as a system wide screen calibration profile with dispcalGUI, the same as you do with dispwin -L. And then I load then same profile to the color managed applications as a screen color profile? It's then I load then profile into then color managed applications that then green cast ampere.

Sounds like the profile is not very accurate. You could try creating
a cLUT base profile using targen/dispread/colprof, and this may
be a better fit. (The Shaper/Matrix profiles are only a good fit
for devices that behave additively. Your experience above hints
that this is not the case with this display. LCD's are notorious for
not so behaving, because the CRT like characteristic is being
synthesised by the electronics.)

Good luck,
    Graeme Gill.

Magnus Berg

E-postadress            mailto://magnus@xxxxxxxxxxx
Hemsida&  Blogg         http://www.magnus.burgsvik.se
Bildarkiv               http://www.gotlandsbilden.se
Gotlandsblogg           http://www.blogg.gotlandsbilden.se
Telefon                 08-519 70 274 - 070-385 87 67
Skype                   magnus_berg,burgsvik
Postadress              Bäckströms väg 47 - 623 35 Burgsvik - SWEDEN

Other related posts: