Hello Graeme On 10-Jan-2011, Graeme Gill wrote: > There is only one possible result of solving it: gamut is traded for > smoothness. > As to how to create a profile that achieves this trade-off, I have quite a > few ideas, but they are non-trivial to implement. I saw there're many research papers around on this topic. For example: http://ee.washington.edu/research/guptalab/publications/2009-CIC-Smooth-LUTsWithAnimation.pdf http://cilab.knu.ac.kr/seminar/Seminar/2009/20090516%20Accuracy-Preserving%20Smoothing%20Of%20Color%20Transformation%20LUTs.pdf but surely you are aware of all them already >> These things shouldn't really happen. > I'm not sure why you say that. Even if inks mixed in a simple way (and > inks of various compositions don't, particularly if they have pigments in > them), the fact that there are multiple ways of creating blacks and > independent ink choices (CMYK) means that it is quite possible. I meant, ideally things like CMYK being lighter than CMY, or even K being lighter than CMY are annoying eveniences. I tried hard-limiting ink channels in my software in order to obtain a more predictable behavior (cmy being a dark grey and still a bit ligther than k alone; 100% cyan being more a cyan than a deep blue, more likely to what happens with regular offset inks, and so on) but the tradeoff is gamut, of course: doing so I totally lose the capability of obtaining saturated secondaries. >> It would be interesting to know, however, how >> other profilers work out this condition. As soon as I have additional time >> I will try PM to see what comes out (I never made plain paper profiles with >> PM). > I have no specific knowledge of how or even if they tackle this problem. I really have to test yet >> The matter, however, is that no artifact should be visible. I will try today >> to see what happens with a profile made out of some 3000 patches and -qh > The nature of the test chart won't changed the fundamentals of how the > inks behaved. This is how I see the problem, definitely. Assume the simple case, when I for testing started with -kx and I looked at what happened. The K channel is turning down at about 90% density, an that's is fine, since I understood why and that made a sense at the end. The problem is, I shouldn't being noticing it visually as a bluish bump or other irregularities. If I notice it, I assume that colprof has not enough sample points (patches) around that zone to work out a reliable result, i.e to resolve this even abrupt transition in a way which is colorimetrically acceptable and without high errors (to the eye, at least). At the end, I could well be WANTING such a K curve, and I assume that if colprof had enough patches in that zone, it could make out an abrupt transition which is however visually precise and without visible discontinuities. I just can't figure out how to accomplish that in a simple fashion. I would have, perhaps, to add more patches in that zone with xicclu, but it would require a dos script at least and I can't play with scripts. So I tried, just for understanding, the following way, but without big luck. Making a first good profile starting with 3000 patches and -qh, so the A2B table (which is the device behavior) is precise enough. The profile is made intentionally with -kx, which gives the abrutp K ramping down. Then I created and printed a second target passing the previous profile with -c and using the -I option (as suggested somewhere in this list) and I made a second profile with it, always using -kx. Thus, I was expecting to see an improvement in handling that area, i.e less noticeable visual discontinuity. Unfortunately that was not the case, or if so, the improving is very very little. I hope I could explain my concern with my poor English (I'm Italian). > >> or register problems). But when the used K is actually weaker than CMY, >> one would have the temptation to simply drop it :-D > Perhaps this is why some printers have "matte black" inks for matte paper ? I never tried the matte black since I heard it actually contains some dyes and may not be durable as pigments. >> I noticed that -K sometimes offers smoother results, but it's not a rule > A specifically chosen -k curve is usually a better approach. And it's not easy. I experienced how a predicted curve looking reasonably smooth may later appear slightly different when making the actual profile. It's almost trial and error. Sometimes the final profile shows discontinuities which were not predicted; sometimes else the opposite happens: the final profile has curves smoother than predicted. But the issue I see is always the same: the problem is not so much how "good" the K curve looks, rather what the visual result is. If, let's say, the transition in the source space from red to black leads to one graysh bump around 80%, how is it that colprof doesn't notice it and try to fix it ? Perhaps those bumps actually originates at interpolation level, i.e between actual B2A grid points ? /&