Hello Gerhard On 22-Jan-2011, Gerhard Fuernkranz wrote: > Reading your objectives, I'm wondering whether it may be a reasonable > approach to resolve the Lab -> CMYK ambigiuties by constraining > smoothness and/or steepness of gradients (considering the derivatives of > all 4 output dimensions in combination, not just smoothing each output > dimension separately), instead of "traditional" black generation approaches? It's hard for me telling how to accomplish this goal in detail. I don't figure out now how such constraining could be done, after all I never tried to write real profiling code in my life and I frankly don't think I will never do :) It was just my suspect that forcing the same black generation curve thru all the tints (not just from white to black but from every Lab=100,x,y color to black), and then computing C,M,Y towards this reference curve (which should not be allowed to drift too much from its shape, or even nothing at all), very better results in terms of coherence (in progression) and smoothness could be achieved. And that would mean almost reducing it to a 3D problem. Unfortunately I can't do everything, I would have made real world tests otherwise. > Furthermore it seems you would also be willing to renounce some gamut > volume, if the larger gamut can only able be achieved at the cost of > steep gradients in the extended region [when limiting the gamut > intentionally, one must take care though, not to introduce holes inside > the gamut volume just because the gradients happen to be too steep or > the curvature is too high in a particular region]. If the potential gamut volume is very irregular, and I at most cut off some bumps, who cares ? The eye is the best judge, not math, after all... That's why some actual test should be made. I wonder if there's some trick [for personal testing purposes only, of course] using link profiles or some other argyll tool [I'm not very expert here...] to create an abstract (idelized ?) CMY->CMYK separation, then profiling the CMY side and then linking the two profiles together... not sure I could explain well /&