Re: assigning profiles
Re: assigning profiles
- Subject: Re: assigning profiles
- From: Graeme Gill <email@hidden>
- Date: Wed, 25 Sep 2002 15:00:01 +1000
Roger Breton wrote:
>
At the risk of taking your point out-of-context, what you describe does not
>
seem to apply to digital camera profiles? Or should it? I mean, I have
>
worked with two different digital camera profiling packages and both don't
>
"have a color conversion that is completely white point relative". When
>
converting from device white (RGB=255,255,255) to Lab, they never give me
>
Lab = 100,0,0.
There are numerous ways one can make profiles, and what is useful,
will depend on how they are going to be used. It seems to be an
implicit assumption of the whole whole ICC profile approach, that
it work most naturally in a relative colorimetric mode (hence
the lut and matrix data being stored in this way, rather
than being stored in absolute form). This translates in
practice to most naturally mapping white in one colorspace,
exactly to white in another. Most users seem to expect this
behaviour too.
This has interesting implication when profiling input devices
like scanners and cameras. For an output device it is easy
to exercise the full gamut that the device is capable of,
simply by using the full range of device values. It is also
very obvious what white is (no colorant on subtractive
devices, maximum colorant on additive devices). For an
input device this is more difficult, because you need a
real world test target as the stimulus, and it is constrained
to the colors achievable in some sort of real world reflective media.
In general, this means that the test chart exercises
only a sub set of the gamut of an input device, and
since the input device doesn't have any inherent white point
(its device value maximums may not be reachable for any
real world test chart), the white point needs to be chosen
somehow.
If the test chart is exactly representative of the media
that will subsequently used, then this is not an issue,
and normalizing the profile to the white point of the
test chart is exactly what one would hope for, so that the
white of the scanned/photographed documents, maps to the
white of whatever colorspace it gets transformed into.
If you are hoping to create a profile for the colorspace
of the scanner or camera, but then use it with input that
is not represented by the test chart you used, then you
have a few problems. One is that you may get inputs outside
the range exercised by the test chart. This means that
you are hoping that the profile can not only interpolate
between the test points it was created from, but can also
usefully extrapolate beyond the test points.
Another implication is that you probably don't want the profile
normalized to the white point of the test media, but
want it to remain an inherently absolute profile
(presumably you will normalize to the white point
after converting to another colorspace, perhaps as
a manual process).
This latter usage doesn't seem to me to be consistent with the
usual assumptions of ICC profiles, and would dismay many users
expectations (their scanner/camera white isn't white), but I'm
not sure what approach most profiling packages take, or whether they
provide a user configurable option to control this.
Graeme Gill.
_______________________________________________
colorsync-users mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/colorsync-users
Do not post admin requests to the list. They will be ignored.