[argyllcms] Re: AW: Re: AW: Re: Question about viewgam
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 26 Feb 2021 09:52:49 +1100
Yves Gauvreau wrote:
just so I understand better, an input device doesn't have a gamut per say but
it as a specific
spectral response (SSFs), right?
Gerhard Fuernkranz wrote:
Keep in mind, that non-colorimetric RGB sensors are colorblind w.r.t. to the
human vision. I.e.
there does not exist an unambiguous RGB -> XYZ mapping which is unique and
correct for all
reflectance spectra. Building the profile for a { scanner, media} pair
instead of for the scanner
alone is eventually the attempt to disambiguate the colorblindness by
constraining the domain of
definition to a small subset of all theoretically possible reflectance
spectra. For some media
types this will work, while for others the ambiguity can only be reduced, but
still not eliminated
completely.
To expand on this, if the input devices spectral response is not a 3
dimensional linear
transform of the average observers spectral response, then it cannot see color
the
same way as humans do. So a profile can attempt to convert the input devices
values
into equivalent human observers colors, but it will be imperfect, and there are
trade-offs that the profile maker has to choose in terms of the amount of error
and
where those errors appear. If you choose a threshold of acceptable error, then
you
can use this to define a "gamut" over which you regard the input device as
having
acceptable performance. But note that this "gamut" is in the high-dimensional
spectral
space, and need not be a contiguous volume - it probably has "holes" in it where
the color error is above your chosen threshold. In general there is certainly
no neat
3 dimensional colorimetric gamut implied by this, and you can't compare it to a
color
output device gamut.
(As Gerhard points out, if you constrain the illuminant and the nature of the
colorants,
then the dimensionality may approach 3, and the spectral errors can then be
much reduced
in the profiles conversion.)
Cheers,
Graeme Gill.
Other related posts: