[argyllcms] Re: Wrong gamma / color rendition worse than before calibration

  • From: Florian Höch <lists+argyllcms@xxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Wed, 07 Sep 2011 21:19:19 +0200

Am 07.09.2011 20:18, schrieb Thomas Bartosik:
Well, that's bad news.
Though I cannot agree on the viewing angle sensitivity (I can barely see any color 
shifting, even at close to 180°...)

A good viewing angle might indicate a PVA or even IPS type panel, but those are seen rarely in mobile computers. A brief search for the T61p panel specs didn't turn up much, so I concluded (maybe wrongly) it's a TN panel.

I think you're the one to know this better...

Not necessarily, as I don't have a T61p :)

My only question is this:
Is the screen only subjectively better without calibration or is the reduced 
gamut a problem for argyll and the result is therefore bad?

Probably the former (my attempt at explanation below).

I mean before calibrating and profiling I could at least differentiate more 
color nuances than afterwards. This is somewhat strange.

Not strange at all: Without the profile, the gamut of everything that is viewed on screen is squashed down to the comparably small gamut of it, leaving nuances intact, but at the expense of saturation.

What IS better after calibrating and profiling?

All colors that are in-gamut will (or should) match better after profiling.

To get some numbers how the screen copes with e.g. sRGB source material, you could use dispcalGUI's profile check feature:
First, select the profile you want to test under "Settings".
Then choose "Verify profile" from the "Tools" menu.
In the file dialog, select "sRGB.icm" as simulation profile.
In the next dialog, select one of the "verify" testcharts.
Finally, you'll be asked to choose a name and location to save the report under. After the measurements, you'll get a graphical report with the color patches, expected vs. measured values, Delta E numbers and bar graphs. It'll probably be quite red (ie. bad) because of the screen's small gamut though.

Even the worst display I have ever seen (also in a notebook, viewing angle of maybe 
60°, and the white level problems mentioned in my first post) got A LOT better 
by calibrating it.
Or is seeing more brightness steps no improvement? I don't know, maybe it's too 

In an uncalibrated state, I get output from my printing company that's pretty 
close to what I design. After calibration I am sure the output will differ a 
lot more. I am questioning whether there are situations where not calibrating 
might be better. Or is this a limitation in argyll?
I understand my display may not be as good as I thought it to be ;-( but 
shouldn't calibrating at least improve it a little and not make it worse?

Are we only talking about calibration (tone curve and white point adjustment), or also in conjunction with the profile (and software that actually uses the profile)?

I have uploaded a HQ pass done in dispcalGUI (just tried it out: The graphs are 
pretty neat!) to

What still makes me wonder is that no other LG panel in a thinkpad has that 
strange strong low end bending of the gradation curves. They are all different 
but more similar to one another that to mine.

Those depend on the tone curve that is chosen upon calibration. Judging from looking at the calibration curves, I'd assume the native response of the screen is maybe closer to 1.9ish than 2.2 in the lower end towards black (but it's obviously only a wild guess on my part).

I somewhere read the green gel used in the spyder2 display filter might give 
wrong measurements over time. Can I completely rule out errors in my 

The filters in most colorimeters degrade over time, so it can't be completely ruled out.

Thank you for your time and explanations!

Florian Höch wrote:


if the profiles are any indication, then the color gamut of that screen
is (a lot) smaller than sRGB (not uncommon for TN panels, especially in
Laptops/Notebooks), leading to clipping.
Same for black. Many Laptop/Notebook displays have a pretty poor black
level (some monitor profilers cheat by scaling the black of the profile
to absolute zero. Argyll does not do this).
I wouldn't pay too much attention to those gamma test images, as TN
panels are very viewing angle sensitive, and combined with the usually
pretty 'bendy' calibration curves needed to make such a display behave,
there isn't much useful information to discern from those images.
In summary, I'm afraid your problems are not fixable via different
calibration/profiling settings, as they seem like hardware limitations
of the screen.

Am 07.09.2011 13:18, schrieb Thomas Bartosik:
Hi list!

I am somehow at a loss as to how to solve my gamma problems..

1) What I want to do:
Have a calibrated display for a) web work and primarily b) print work
sent to a printing studio. They accept ISO coated CMYK PDFs.

2) What I have done:
I got a spyder2 for cheap, i know it's not the best, but I'd like you to
comment on the issue and tell me whether the colorimeter's quality might
be the problem.

I used the colorimeter to calibrate/profile really bad and cheap
displays (e.g. one notebook display that did not differentiate between 
white till 255,255,255 white, everything was white)
The calibrated/profiled results of these displays are very pleasing and
now show nuances that were completely unseen before.

Now my problem is my own display in a Thinkpad T61p. This is (an
expensive) LG 1920x1200 display that has really good reviews in most articles I
have read so far. In an uncalibrated/unprofiled state, colors look really
good but a bit bluish and the brightness distibution is also very good
(subjectively, to my eyes). i.e. I can differentiate 252,252,252 white from full
white and also 3,3,3 black from full black. This has not been the case with
nearly all other uncalibrated/unprofiled displays I have seen so far.

I did several dispcal runs, and they more or less ended up in the same
color rendition.
(I tried with -t 6504 and -gs and without both)

Now the problem I face is this:
All dark nuances in pictures are too dark. And I think I can really say
that in an absolute manner, it's not only a subjective thing.
The same goes for bright tones.

If I look at
in a color managed app like gqview or gimp with the .icc profile loaded
and applied, I cannot differentiate between FF66FF, FF33FF and FF00FF while
I can do so unmanaged or with just dispwin loading the profile and the app
not loading it. As far as I have read till now this huge difference
between using a color managed desktop (i.e. loading the profile with dispwin or
xcalib) and then applying the same profile again with a color managed app
should not be that dramatic.
I suspect something's wrong with the measurements taken or the profile
Can anyone give advice as to in which direction I should look to solve
this problem?

As an interesting sidenote, I found a profile for a similar display ( A
T61 LG panel, resolution could be different, I don't know more about this
profile), and apart from its D50 whitepoint the effect of not seeing
different magenta nuances is quite the same (at least FF66FF and FF33FF are
indistiguishable). This profile was taken with a gretag macbeth device, I guess
on windows.
I found an additional profile that's also for an LG in a Thinkpad, with
the same effect on magenta.
This makes me believe my profiling data and measurement values seem to
be probable at least. But why don't I see those magenta nuances?(the same is
also valid for some green and blue areas...)

All of the profiles and my cal and icc file are at
(The T61p-full-brightness... is mine)

Do you see differences in these colors?
If I do the monitor test on
after having applied a profile, I can barely make out the lowest stripes
in the contrast test. The gamma test seems more accurate than without
being calibrated, but the area of blending in is a bit broader. The bars all
blend it at around 2.2 which I never managed to do with just the nvidia
controls, but the blue gamma is somewhere around 2.6, which puzzles me.
NB: This is when I test it in opera, i.e. the .icc profile is loaded by
dispwin but not the browser!
If I download the contrast and gamma .png and look at them at zoom 1:1
in gqview with the profile loaded, I get these gamma readings:
          48%     |     25%      |     10%
   R   G   B   W  |   G     W    |   G    W
2.3 1.6 2.8 2.3 |  2.2   2.3   |  2.2   2.3

The 1.6 at green is pretty extreme, and the 2.8 at blue as well (tho I
cannot really say it is 2.8 for sure as 2.5 till 3.0 is nearly the same
level to my eye)

Without any calibration, the gamma and contrast test are perfectly OK,
but could be a bit better (this is what I wanted the calibration/profiling
to do for me!)

If I tune the values in nvidia-settings for the lagom lcd test and dump
the values with dispwin -s I get nvidia.cal (also at the webspace above).
This is the (subjectively) best setting I have till now...

Maybe I have to give different -g or even -G values? I do not understand
-G at all and I think -g should best be set to the display's native gamma
as it will result in the least banding artifacts and does not really matter
as the color managed app should compensate for it anyway...

I hope there's some advice you can give as I am somewhat clueless..

Thanks in advance!


Here's a rundown of the calibration I've done.

blackknight ~ # dispcal -yl -v -R
XRandR 1.2 is faulty - falling back to older extensions
Setting up the instrument
Instrument Type:   ColorVision Spyder2
Serial Number:     00633553
Hardware version:  0x0307
Place instrument on test window.
Hit Esc or Q to give up, any other key to continue:
patch 3 of 3
Measuring VideoLUT table entry precision.
patch 6 of 6
patch 6 of 6
patch 9 of 9
patch 9 of 9
Uncalibrated response:
Black level = 0.40 cd/m^2
White level = 119.19 cd/m^2
Aprox. gamma = 2.29
Contrast ratio = 300:1
White chromaticity coordinates 0.3269, 0.3578
White    Correlated Color Temperature = 5726K, DE 2K to locus = 12.5
White Correlated Daylight Temperature = 5726K, DE 2K to locus =  9.6
White        Visual Color Temperature = 5366K, DE 2K to locus = 12.2
White     Visual Daylight Temperature = 5475K, DE 2K to locus =  9.3
Effective LUT entry depth seems to be 10 bits
The instrument can be removed from the screen.

blackknight ~ # dispcal -m -yl -t 6504 -gs -v -O "Thinkpad T61p
1920x1200 LG full brightness Argyll-1.3.1 sRGB D65 Spyder2" -o
/usr/share/color/icc/T61p-full-brightness-argyll-1.3.1-sRGB-D65XRandR 1.2 is 
faulty - falling
back to older extensions
Setting up the instrument
Instrument Type:   ColorVision Spyder2
Serial Number:     00633553
Hardware version:  0x0307
Place instrument on test window.
Hit Esc or Q to give up, any other key to continue:
Display type is LCD
Target white = 6504.000000 degrees kelvin Daylight spectrum
Target white brightness = native brightness
Target black brightness = native brightness
Target gamma = sRGB curve
Commencing device calibration
patch 6 of 6
Black = XYZ   0.35   0.35   0.40
Red   = XYZ  44.06  26.35   3.27
Green = XYZ  33.71  59.85  10.98
Blue  = XYZ  17.84  17.99  78.99
White = XYZ  95.06 103.78  92.64
patch 128 of 128
Initial native brightness target = 103.784294 cd/m^2
Had to scale brightness from 103.784294 to 81.568913 to fit within
corresponding to RGB 0.899985 0.837122 1.000000
Target white value is XYZ 77.528580 81.568913 88.828359
Adjusted target black XYZ 0.34 0.35 0.40, Lab 3.88 0.45 -0.30
Target black after min adjust: XYZ 0.342 0.351 0.399, Lab 3.883 0.447
Gamma curve input offset = 0.000000, output offset = 0.004299, power =
Total Iteration 3, Final Samples = 64 Final Repeat threshold = 0.600000
Creating initial calibration curves...
Doing iteration 1 with 16 sample points and repeat threshold of 1.200000
patch 16 of 16
Brightness error = -0.462568 cd/m^2 (is 81.106344, should be 81.568913)
White point error = 0.625266 deltaE
Maximum neutral error (@ 0.895486) = 1.116106 deltaE
Average neutral error = 0.685531 deltaE
Number of measurements taken = 26
Computing update to calibration curves...
Doing iteration 2 with 32 sample points and repeat threshold of 0.848528
patch 32 of 32
Brightness error = -0.322559 cd/m^2 (is 81.246354, should be 81.568913)
White point error = 0.646947 deltaE
Maximum neutral error (@ 0.248961) = 0.841565 deltaE
Average neutral error = 0.534482 deltaE
Number of measurements taken = 49
Computing update to calibration curves...
Doing iteration 3 with 64 sample points and repeat threshold of 0.600000
patch 64 of 64
Brightness error = -0.351730 cd/m^2 (is 81.217183, should be 81.568913)
White point error = 0.201203 deltaE
Maximum neutral error (@ 0.100647) = 0.920412 deltaE
Average neutral error = 0.409086 deltaE
Failed to meet target 0.600000 delta E, got worst case 0.559542
Number of measurements taken = 159
The instrument can be removed from the screen.
Written calibration file
Luminance XYZ = 0.000000 79.468347 0.000000
White point XYZ = 0.951621 1.000000 1.047750
Black point XYZ = 0.004093 0.004148 0.004938
Created fast shaper/matrix profile

blackknight ~ # dispcal -yl -v -r
XRandR 1.2 is faulty - falling back to older extensions
Setting up the instrument
Instrument Type:   ColorVision Spyder2
Serial Number:     00633553
Hardware version:  0x0307
Place instrument on test window.
Hit Esc or Q to give up, any other key to continue:
patch 3 of 3
Current calibration response:
Black level = 0.41 cd/m^2
White level = 80.85 cd/m^2
Aprox. gamma = 2.20
Contrast ratio = 196:1
White chromaticity coordinates 0.3122, 0.3277
White    Correlated Color Temperature = 6541K, DE 2K to locus =  4.0
White Correlated Daylight Temperature = 6542K, DE 2K to locus =  0.7
White        Visual Color Temperature = 6395K, DE 2K to locus =  3.9
White     Visual Daylight Temperature = 6566K, DE 2K to locus =  0.7
The instrument can be removed from the screen.

Florian Höch

Florian Höch

Other related posts: