Elena [service address] wrote:
See my other mail also. And please forgive me if I sound repetitive or just chaotic =)
Thanks for your suggestions.
Looking things from the point of view of a non-experienced or little-experienced user (experienced here is relative to argyll skill, not to CM in general!) - one would expect a correct result already with some default parameter - and correct here doesn't mean perfect, just overall correct and without major visual issues.
Most of the time, this is the case. I think that you have just started with a particular device/paper/ink combination that highlights the current shortcomings.
Now, when an user makes a choice on the K generation, usually this choice is driven by factors such 1) I prefer a moderate or almost null K in the light tones because I don't want to see rosettes (or pepper) too early 2) I use a fine screen so I want my neutrals being composed as most as possible by black, so I minimize metamerism issues 3) I prefer a purer K at deep blacks rather than a heavily composed one because my paper behaves better this way - EVEN IF THAT WON'T PERHAPS RESULT IN MAXIMUM (measured) DEPTH, an aware choice, etc, etc, etc.
Sounds reasonable.
One turns out instead to have to play heavily with xicclu just for eliminating visual bumps or irregularities.
It depends heavily on the device/paper/ink. Many devices give very good results without great sensitivity to the black curve.
Now, don't just consider my experience with plain paper any longer. Those or even worse things do happen, with many different substrates. Think of people who need to profile solvent printers, or to print on aluminium! As I already said in the other msg, I think colprof should be wise enough to detect major visual irregularities, and in the case, suggest a number of patches to add to a new target trying to correct the problem. This could be a file automatically generated, to be integrated in the previous ti1 target. I'm telling that, of course, without having never written profiling code, of course, so be tolerant... :-)
If it were just a matter of the number of patches, then this would be reasonable. But if it's really a problem with trading off gamut size for K smoothness, then I have no means currently of addressing this, since fundamental changes to the code are needed. Even if there was a means of automatically solving the problem, you would find that this device/paper/ink combination has a poorer gamut than ideal, since some of it would have to be avoided to avoid the black smoothness problems.
What I suggest, is that -k parameters should be taken literally, within the total responsability of the user, even if that won't perhaps lead to 100% accurate results (note that one thing is accuracy, while smoothness and continuity is another thing).
It's fundamental to the current code, that accuracy has priority over desired K level. I don't think that simply having this the other way around would really work, as it would then be a return to "open loop" color, rather than measured color. It's possible perhaps to better smooth the K generation curve to the K values at the gamut edge, but this wouldn't solve the problem of the gamut edge itself containing sudden changes of K value. The ultimate solution is to trade gamut for K level smoothness/consistency, and in that sense make K a priority over accuracy. Graeme Gill.