[argyllcms] Re: Displaying sRGB graphics on wide gamut monitors - gamma problem?
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Thu, 28 Oct 2010 00:08:42 +1100
Anders Torger wrote:
As far as I understand - apart from black point compensation, perceptual
intent should be the same as relative intent for sRGB on a wide gamut
This is not necessarily true. A cLUT based profile can have quite different
gamut mappings for perceptual and relative colorimetric.
A matrix profile will have no difference.
screen, and since I have calibrated to D65 it should be same as absolute
too, since there's a match of white point and sRGB fits within the
gamut. However, concerning gamma there might be differences... I could
imagine that perceptual intent is looser on gamma correction than
relative or absolute, if there is a standard for it.
Beware of drawing too many conclusions about absolute colorimetric
and display profiles. While Argyll is perfectly self consistent
within itself, other ICC V2 profiles and CMM's may do things differently,
due to different interpretations of the ICC specifications.
Some display profiles assume that the displays native white point
is an "illuminant", and that it should be chromatically adapted
to D50 before profiling. This results in an absolute colorimetric
intent that is the same as relative colorimetric, and leaves no
mechanism for recovering the actual display white point using
intent selection. ICC V4 clarified the specifications to make
this the "correct" approach.
However, as I understand it how rendering intents is actually rendered
by applications is not strictly standardized. For example if make a
perceptual renderer and shall show an sRGB image on a 2.4 gamma screen,
you may choose to not correcting for that. I'm not yet sure how this is
done, however my guess is that most perceptual renderers don't correct
gamma, but perhaps that's wrong? Is rendering intents strongly
standardized so one can be sure that all applications renders the same
way? Apart from gamma, it seems like black point compensation can be
made in several different ways and there's no standard for that either?
Black point compensation can be seen as an attempt to do a partial
(that is luminance only), on the fly, gamut mapping. This is
to compensate for the static nature of the gamut mappings baked into
ICC profiles. A true gamut mapping takes into account all aspects of
the source and destination gamuts, including the luminance range.
The gamma is irrelevant as far as a CMM is concerned, since gamma
refers to a relationship between the device values and response,
while the whole point of color management is to make the reproduction
device value independent. To put this concretely, you can calibrate
your display to any gamma you like, and it should have no direct
effect on the display of a color managed image on the screen,
because the display profile will measure the response of the
display, gamma and all.
If the renderer would correct for the monitors gamma, I still have a
problem. I don't actually want 1.0 viewing gamma, since my ambient
lighting is low. So somehow the renderer would need to know how to
correct for that... but perhaps I can encode my viewing conditions in
the monitor's ICC-profile, and then good perceptual renderers will
adapt?
Calibrating a screen has the largest effect on un-managed colors
(ie. colors defined in display device RGB values). It does set
the white point and brightness, but that is all when it comes to
color managed display. The viewing conditions can be taken into
affect in the gamut mapping, if the gamut mapping is told
about the viewing conditions of the source and destination.
So far my assumption is that applications cannot really be trusted on
what they do, so it is best to do most calibration in dispcal (where you
actually can see what is done) and not leave so much need for correction
in the applications.
Please take some time to understand the difference between calibration
and profiling.
The experiment of not calibrating at all, just doing characterization
and let applications do all correction and compare that with calibration
was an eye-opener to me. As a layman one would think that there would be
no difference, but there really is a huge difference in how things look.
I currently would guess that there may be huge differences in how gamma
is handled and how black point (and white point) compensation is made.
But I seem to have lot to learn yet. Knowing what one *actually* sees
surely does not seem to be an easy thing...
If you are seeing huge differences between profiling the uncalibrated
vs. calibrated display and then displaying color managed images
(apart from a chosen white point and brightness change), then something
is going horribly wrong. You need to figure out what.
Graeme Gill.
Other related posts:
