Graeme Gill wrote:
Gerhard Fuernkranz wrote:I guess it is not completely WP relative be cause of the incomplete adaptation (D factor != 1) in the CIECAM model? Might it make sense to force D=1, instead of computing it from the viewing conditions, for intents which are supposed to map src WP -> dst WP? Would this probably result in perfectly aligned WPs in Jab space, without need for further alignment?may I ask, how do you actually align the two white points (and how do you adapt all other source colors according to the WP shift)? Just a translation in Jab space? Chromatic adaptation? Rotation from src WP to dst WP in Jab space with center at [0,0,0]? Etc.?Currently it's a rotation in Jab space about 0,0,0. Usually the shift is small, because Appearance space is close to being white point relative.
I fully agree, that a skewed gray axis from the device WP to a non-neutral BP is rather not desired (and as you know I was also one of the people who did complain about this behaviour on previous versions). But why is it better to inherit the gray axis direction from a given source profile? If the source profile is a working space like sRGB, then it's gray axis is indeed not skewed, but what if the src profile already has a skewed gray axis from WP to a non-neutral BP? Wouldn't then the device profile I'm generating again inherit a skewed gray axis?So if I understand you correctly, the direction of the perceptual gray axis of the device being profiled is eventually determined by the wtpt..bkpt axis of the supplied _source_ profile (-> profile ... -S ...), while the lightness of the lower end of the device's gray axis is determined by the device's bkpt? What's actually the intention of this strategy? Looks like the created device profile inherits some of the source profile's personality (gray axis direction), but is this really desired in general? Shouldn't the device profile being created for a particular device rather represent the "native personality" of the device?From various feedback, I understood that this behaviour is not desirable, in it's shifting of neutrals towards the hue of the black point.
Regards, Gerhard
This is consistent with the findings from Sun Ju Park and Mark D. fairchilds paper "Color Reproduction Using Black-Point Adaptation" <http://www.cis.rit.edu/fairchild/PDFs/PRO16.pdf>, which found that "Color reproduction based on being relative to the black and white in a complete sense using chromatic and lightness values is not preferred"
Subjectively, the algorithm I've adopted gives results that are a closer visual match through the neutral axis, than the previous one.
Graeme Gill.
