[argyllcms] Re: xicclu -g predictability issue

  • From: Elena [service address] <1007140@xxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 31 Jan 2011 14:14:11 +0100

Hello Gerhard

On 31-Jan-2011, Gerhard Fuernkranz wrote:

> My understanding is that "value" refers to an actual amount of K, while
> "k locus" refers to a relative amount of K (where locus =1.0 refers to
> the K value that would be selected by -kx, while 0.0 refers to the K
> value that would be selected by -kz).

That apparently was my understanding also, but it turns out a nonsense,
since then a required -k separation should be maintained thru all the
gamut irrespectively whether some colors can't be faithfully reproduced
with the required black, while a -K one should be allowed to change
for a faithful reproduction, even at expenses of discontinuities. But
that seems not the case (as I already discussed weeks ago)

> "-kz" attempts to reproduce a given L*a*b* color exactly, using as
> little K as possible. If it happens that the given color can be
> reproduced without K, then consequently no K will be used at all. For
> colors which cannot be reproduced without adding some black ink
> (typically rather darker colors, near the gamut boundary), -kz does well
> generate non-zero K.

but as I shown with my latest -kz plot, there're areas where the minimum
K needed rises up (or even falls down) abrutply :-/

> I don't know yet whether this is really just noise or anything else. One
> would likely need to investigate all the involved cells of the A2B grid
> in detail - but that's quite tedious...

Yes, I don't know where and how to start such an analysis.
Assume you isolate one discontinue area. You then should be able to
find out all CMYK combinations (how ?) having the Lab color of that area,
print them and integrate the measurement in the original .ti3, to see
if the bifurcation is real (then colprof should really be able to chose
alternate branches) or it disappears :-|

But let's assume that to shape out the real printer gamut one million
patches would be required, also printed in different locations and
each reading averaged 2/3 times. Something unpratical. So, better
having the inversion routines being able to detect discontinuities
and to chose alternate branches. Paranoically following the gamut
shape is the wrong way to go, because in the real world it will never
be so discontinue as it appears. I think so.


Other related posts: