[argyllcms] Re: Calibration issues using argyll

Thanks for your detailed answer Peter. I will try to be as exact as possible in my reply. 

Peter Karp schrieb:

iColor Display was tuned to give a visual good gray balance.
Especially for the hardware calibrated Quato displays, which is the
"main task" for iCD, it does a good job.

You have not specified which blackpoint setting you have choosen for
the iColor Display calibration. There is one setting to calibrate to
the native blackpoint (meaning as dark as possible) or you can choose
to calibrate to a specific blackpoint luminance with "neutral black".
The latter is the setting I use, neutral black with a target value of 0.0 cd/m^2. I think what it does then is try to achieve a black luminosity as low as possible without sacrificing neutral rendering. The results are quite good. The black point is measured to be about 0.11 cd/m^2, which is virtually no different to the black point luminosity without any calibration. 


This setting will use the specified blackpoint and will try to match
the blackpoints color temperature as close as possible to the
whitepoint. This is similar to choose the "blending" in Argyll. But as
Graeme said this is a trade off between a low black point and a more
or less neutral color temperature in the shadows. In my experience a
totally corrected black point is not needed and will result in a too
high black point. "Too high" means here you will loose contrast. You
won't see 1000K higher color temperature in the 0/0/0 black, when a
little above (for non-zero RGB values) you will get close to the white
point color temperature. You could upload or e-mail the generated
profiles and other data files. That might be helpful to find the
reason for your observations.
I will put links and descriptions to the various created profiles at the end of this message. I just took some readings with spotread using a canvas in Photoshop as a "patch". I will explain my procedure shortly: First, I filled the canvas with pure white and took a reading then to be used as the white point reference (Shift + a). Then, I filled the canvas with black and gradualy increased the luminosity in LAB space in steps of 10. The readings (Button a) are as follows (temperature values CCT): 

000: 7838K
001: 5340K
002: 5570K
003: 6264K
004: 6391K
005: 6533K
010: 6257K
020: 6389K
030: 6396K
040: 6495K
050: 6420K
060: 6455K
080: 6513K
100: 6475K
I have to say that readings below a L value of 5 were quite inconsistent. For example, the black point (L=a=b=0) was measured as 
#1 XYZ: 0.091332 0.107938 0.132846 (08952K)
#2 XYZ: 0.091332 0.091332 0.141149 (10218K)
#3 XYZ: 0.099635 0.107938 0.132846 (07896K)
These are three successive measurements. Of course, there are just very slight variations in the XYZ values but these result in huge differences in color temperature. 


The calibration matrix used should not make a difference. Your Eizo
uses a "sRGB"-panel type and iColor Display uses the LCD readings from
the SDK. In this case no special calibration matrix is used, but the
default X-Rite factory one for LCDs. I assume that Argyll also uses
the LCD values and neither the raw measuremet data nor the CRT values.
So the instrument readings should be similar.


That would also be my impression. Regarding the UGRA-Test you wrote:


It has some relevance but not in this regard you're trying to
understand. It's a closed loop test and therefore the device will not
tell you "absolute" truth you are looking for here. The test is useful
to some extent, but because you normally won't know if your
colorimeter the use is limted.
This is also what I thought, the test being a loop. Of course you can only proof that the profile created corresponds to the readings taken before and to what extend it is accurate in creating the gamut description. 


I tried the download, but it didn't work (any longer?).

On a related note: One idea I had and always wondered why nobody
seemed to realize (at least AFAIK) is the option to specify for a so
called software calibration (via vcgt) if the calibration curves
should be smoothed, thus giving smoother gradients sacrificing a
little bit of the profile accuracy.
Here is the link to the updated (and fixed) dispcal.exe: http://www.argyllcms.com/dispcal_win32.zip 


After I created a profile, I normaly check out the results by
browsing to two different locations on the web. The first one [1]
renders a test image in which numbers are displayed with different
neutral RGB values. Profiles generated with iColor produce results
where all numbers are visible and none of them seems to have any
color cast to the naked eye.


You have to be extremly careful what you check with such a test
picture. Does it use "raw" values or are ICC profiles used for the
test picture and the monitor? Second you need to know that if you look
if certain steps will be visible in a ramp is highly dependent on the
ambient _and_ the gradation you used for the calibration. There's not
a single "best" gradation.
The images I use have no embedded profile and are only interpreted through the installed monitor profile. I think Photoshop does not do any space conversions except for the one into the monitor space. As I understand it, this is the way I am able to see if a profile created shows any unwanted behaviour visible to the naked eye. Of course, this is a very limited test and not objective at all.
My settings for observer and ambient are the default ones used by Argyll. I ave yet to see a profile created with Argyll or iColor Display that would not show all numbers, so there has been no problem as of now. 


Profiles created with Argyll also reveal every number, but there is
a slight greenish or reddish hue on numbers on the dark area from 4
downwards.


That's the interesting point here. Without having compared both
solutions I guess the reason is the "resolution" (means number of
measurements) in the shadow range which is used as the basis for the
calibration (vcgt) curves.
I normally set the quality to high when using Argyll to calibrate the display which would only allow for a deviation of 0.6dE to the target value. I don't know how efficient Argyll does the job, but looking at the number of measurements taken alone should give Argyll an advantage in accuracy over iColor. 


The second image I open is to see how many visible bands in a
supposedly smooth gradient appear. Here, both programs produce
profiles that show banding


Which you will always have when you have a "software" calibration
opposed to a hardware calibrated display. That's because of the
limited 8Bit/channels resolution for the vcgt "calibration". You have
an 8Bit/channels input signal, do some corrections in 8Bit/channels
for the output. Then you will loose some levels due the inevitable
roundoff errors. Smoothing the calibration curves should improve this
to some extent. For example the default vcgt-curves for Apple displays
(default ICC profile for Powerbook for example) are smoothed and
produce a decent smooth gradient IMO. I don't know if Argyll supports
such a smoothing. But you could do this "by hand": read out the vcgt
curves (xcalib or X-Rites CalibrationTester might be helpful but I
think in the Argyll files you also can see the calibration curves),
import the values to Excel or somewhere else, smooth the function (or
better said values representing the transfer function) and create a
new profile including the smoothed calibration.
I understand that software calibration will always result in banding. But there are differences in various profiles created and I merely check if banding grows out of acceptable boundaries. I did and do not expect Argyll or iColor to create profiles without any banding at all. I know this is impossible with the current 8-Bit limitation and could only be done using hardware calibration (which Argyll does not support at all, as far as I know). I wonder if there are any plans in the industry to broaden the connection between graphic cards and displays to at least 10 or better yet 12 Bit per Channel. I sure hope so for the future. 

Thanks for your time, I hope I could resolve any misunderstandings.

Steffen


Attached are links to the profiles and a quick explanation for each:
[1] http://www.evilstiefel.de/pics/argyll/20080904_Eizo_Native_LinearVCGT.icm This is a profile created by Argyll skipping the dispcal utility and just profiling my screen using targen to create 2000 test patches, reading them in with dispread and using colprof to create the profile. 

[2] http://www.evilstiefel.de/pics/argyll/20080909_Eizo_ProPhotoRGB.icm
This profile was created using iColor Display in reference mode using the recommendations for the ProPhotoRGB color space (6500K, Gamma 2.2) 

[3] http://www.evilstiefel.de/pics/argyll/20080910_Eizo_22_Native.icm
The profile currently in use. I created it with Argyll, set to native white point (which is around 6480K) and a Gamma of 2.2. Creating a profile with a Gamma of 2.4 resulted in annoying gradient results in applications that don't support color management (of course this was just a matter of personal taste, but nevertheless). The new parameter A in dispcal was set to 3 and this time I chose 1000 test patches to be read in (with virtually no difference in accuracy to 2000 patches).

Other related posts: