[argyllcms] Re: icclink -G and source gamuts /-profiles

Klaus Karcher wrote:
This hack applies to XYZ profiles in general? And it has nothing to do with profile's -ni and -no options or icclink's -n option? (I must confess that I didn't comprehend their role as yet). 

At the moment, yes. For an XYZ profile the equivalent of
a "-ni" or "-no" is done in icclink and cctiff.

Actually, on thinking about it I've decided that this
particular problem isn't too hard to fix without
generating optimized per channel curves on the fly.
What I will do is change icclink and cctiff to
add in Y to L* curves between the profile per channel
curves and the CLUT for Matrix (XYZ PCS) profile.
This effectively forces the CLUT input and
output space to an L* characteristic, which for
general transformations should be close to optimal.
This approach made a huge difference to
XYZ CLUT profile generation, so should work well
in this very similar situation.

XYZ CLUT profiles will just have to fend for themselves,
but this is pretty much the situation anyway - if the
input curves for them don't distribute the grid well,
there's nothing much you can do after the fact to
fix it, and if they do, it will work well on
the input of a device link CLUT.

Bumping up the grid res. will help a bit with
fidelity without this, but won't easily compensate
for the grid spacing perceptual non-uniformity.

Graeme Gill.

Other related posts: