Gerhard Fuernkranz wrote:
Using a profile created from Elena's ofps2.ti3 data I did take a closer look at a short line crossing one of the discontinuities (a straight line in the source sRGB space, which is then converted with perceptual intent to CMYK). And it seems to happen at a location where CMYK colorants are crossing very steeply. Note that the 101 CMYK points below spawn only a couple of dE in total!
Right. So it seems that this is not a bug in the inversion algorithm, but reflects the sudden shift in K vs. CMY at that particular spot on the gamut boundary.
So possibly limiting the 1st derivatives may be a key to success?
The problem is that gamut has a priority that trumps any secondary factor. A more complicated approach is needed to set a trade-off between gamut volume and device value smoothness. Graeme Gill.