Hi Graeme, Thanks a million for struggling with my half-baked ideas. Graeme Gill wrote:
Hi Klaus, I notice that the XYZ cLUT extrapolation is better than Lab, but not quite as smooth (perhaps the smoothness needs bumping up for XYZ cLUTs for some reason.)
I haven't tried this yet. I had some sobering experience with XYZ CLUT profiles (distracting artifacts in spite of low self-fit errors), but maybe I should revisit them again.
I had a fiddle with the code, and this is the result: <http://www.argyllcms.com/Argyll_dev_src.zip>.
For cLUT input profiles with the -u flag, it creates a device black and white point, with PCS value derived from a gamma/matrix model. See what you think.
I've compared both version with another data set (same sensor and light source as before, but L*-like TRC instead of gamma 1.5), see http://digitalproof.info/argyll/virtualScanner/center2cusps.pdf>
My observations:* I was surprised to see significant differences also in the gray->cyan gradient
* I was surprised how straight the curves to the colored cusps were in both cases (didn't yet examine where the training set gamut boundary is located and the extrapolation starts)
* The most significant difference between the profiles is in the gray-black gradient (as expected). Above all, the a* curve becomes much better with 1.3.1 beta, but the L* curve get sightly worse.