Klaus Kompatscher wrote:
I get a scanner profile (colprof) with too high DE >40.
Hmm. I'm not quite able to reproduce that. Using the default cLUT type profile, I get a worst case error of 16, average 3.5, which isn't fabulous, but doesn't seem inconceivable either.
The procedure: - I create my own target with ./targen -v -d 3 -s 21 -g 21 -f 513 -c Cruse_LRFB.icc Scanner (ti1 file attached) by using a previous icc profile of the same scanner.
Why are you using the scanner profile to create the print chart ? You want the print chart to have well spread colors in the available printer gamut, so you should use the printer profile for this. Ideally you should opt for full perceptual distribution and no curvature distribution (because you aren't interested in mapping the printers behaviour), so I'd recommend using the (undocumented) options "-A p1.0 -A c0.0" instead of the default or the combined -A adaptation value.
I double checked all data and it looks good to me, nevertheless the profile is unusable.
It's hard to tell what could cause such problems. I would suspect the reference values, but there is nothing as obvious as completely scrambled sample values, none the less the sorts of errors I get for a gamma/matrix profile and the distribution of the fit errors suggests something is not right with either the reference values, or the scanner .tif file (channel values being clipped ?) Graeme Gill.