[argyllcms] Re: Version 0.53 - Gamut mapping warning message.

Roger Breton wrote:

When building the profile with argyll would it be possible that you help us
"peek" at the underlying device behavior with some smoothness statistics?
Like, your code makes some decisions based on the measurements but it's all
happening to some extent inside a black box, even though you supply us with
command line arguments to control the mapping. I'd really like to see what
those non-monotonics statistics look like as that would surely help
appreciate pinpoint errors.


Roger,
        I'm not sure how useful such things would be.
I don't really know how to interpret such internal "noise"
myself, most of the time.

Merely detecting non-monotonicty in a multi-dimensional
mapping is not a well established thing, and the simple
method I'm employing has arbitrary "magic" numbers associated
with it, that set a detection threshold. It's something I put
in the code early on, thinking it might be useful, but
in the end, hasn't been used for anything serious.

How such a metric relates to anything resembling color
appearance, I have little idea. It sounds like a major
project (several months work at least), to establish a
correlation between any smoothness metric and the
gamut mapping behaviour, and then the gamut mapping
behaviour and the subjective results. (ie.
any such metric needs to be "calibrated" to be
useful as a measure of things.) All I did in tweaking
things was check a couple of standard test charts through
the RGB->CMYK path, and look at the RGB gamut surface
color mappings.

Really, anyone that interested at that level in the
internals is effectively doing the research themselves,
so I expect they will be compiling from the source,
and can add reports to things like this themselves ! :-) :-)

Now, if you'd really like a "smoothness" control in the gamut
mapping, that sounds like a more reasonable idea. It's still going
to end up being the users problem to figure out what non-default
setting suite their tastes though.

cheers,

Graeme Gill.


Other related posts: