[argyllcms] Re: Generating new black curve from existing profile.

me@xxxxxxxxxxxxxxxxx wrote: 
Problem:  When plotted on xicclu the curve looks good until about 90% of the
way across (about 15 or so on the lower scale - I assume these are L values?)
When it gets up into the deepest blacks it goes completely haywire, the black
flattens out and the other components start "scribbling" around.


This is pretty normal with the current way Argyll handles CMYK.
There are a number of factors interacting -

 The target output color.
 The target black curve you are specifying.
 The total ink limit.

The total ink limit triggers two effects:

  Typically there will be a point where the target CMYK value
  needed to create the target output color and comply with the
  black curve exceeds the total ink limit. At that point the black
  curve will be deviated from to meet the total ink limit, by
  increasing K and reducing CMY.

  Typically there will be a point where the target CMYK value
  needed to create a target output color exceeds the total ink limit
  even when maximum K is being used. Such points are out of gamut,
  and the that point the closest color that meets the total ink limit
  will be used.

xicclu makes things look worse than they actually will be, since it plots
the L*a*b* -> CMYK lookup on a point by point basis, whereas the real
profile only has such points at it's grid lattice locations, and intermediate
points are linearly interpolated between, smoothing the characteristic somewhat.

[The strange behaviour is because the device characteristic is contained
within a 4 dimensional cube, that's mapping into 3D space. At the light
end, the mapping is uncomplicated, but at the dark end you have 9 of
the 16 cube corners crammed into a small space, forming a complicated,
interpenetrating geometry. The CMYK inversion uses is being done on a
point by point basis, so it tends to jump around between different solutions.]

It's best to play with the black curve and at least try and make it
match up with the first "trigger" point I mention above (hence the
advice in <http://www.argyllcms.com/doc/Scenarios.html#PP6> of
looking at the curve you get with minimum and maximum black).

Also a lot of this is happening at quite dark colors, and the strangeness is
not always visible, depending on the nature of the device.

A project that I've been working on (rather slowly :-) is to change
things to add a smoothness criteria, to make it a bit easier to
use, and hopefully produce a smoother result.

Graeme Gill.

Other related posts: