Hello, Graeme. Thank You for Your attention. Is it possible to detect the "rise to a dip" behavior in profiling software itself? May be there is an automatic work-around method possible for such devices? I can imagine that the profiler performs the test like mine automatically - inverse then forward conversion of a grey gradient and the checking the resulting gradient for monotonicity. If non-monotonicity is detected, some work-around can be performed while profile re-computing. I think the gray scale monotonicity is the most important and it's will be acceptable to check only greyscale inversion. Mon, 30 Apr 2012 15:41:04 +1000 Graeme Gill wrote: > Nikolay Pokhilchenko wrote: > > Can You take a look on the device (TI3 attached in archieve)? > > Perceptual and Saturation intents in ArgyllCMS profile are OK, but relative > > intent is quite > > strange: there is lightness inversion at the black side of a gray gradient. > > Hi, > thanks for drawing this problem to my attention. > It appears to be due to the way that out of gamut clipping transitions from > exact using L*a*b* within the gamut, to CIECAM02 out of gamut. The crossover > region (4 DE) sometimes gives rise to a dip, mainly because CIECAM02 > models flare which compresses the dark region behaviour. > > If I reduce the size of the transition region then this effect is > much reduced, although it does not remove other effects such > as cLUT interpolation artefacts, or other artefacts caused by the > devices darkest point not being neutral. > > [This change will be in the next release. If you're compiling from > source you could change xicc/xlut.c line 1348 from > bf = cdist/4.0; > to > bf = cdist/1.0; > ] > > Note as well, that perceptual intent returns the closest point > in the gamut, so there are no guarantees about monotonicity > or neutrality. If a gamut surface is convex (and many printer > gamuts are convex in places), then the closest point may "jump" > from one point on the surface to another as you move the target > color smoothly away from the surface. Such jumps can't be > represented in a B2A cLUT, and the resultant device value > interpolation may behave non-intuitively. > > Graeme Gill. > > >