Gerhard Fuernkranz wrote: > if you "search" the RGB cube anyway (i.e. adjust RGB, measure, adjust RGB, > measure, ... until > optimum is found), what role does the model play then? For approximation of > the derivatives? (if > the search/optimization is guided by gradients) Hi, I could do that, but I'm not sure how slow it would be. It's unclear how rewarding this would be in general (ie. is it worth the effort ?). > Btw, one could possibly also consider a 2D search here, i.e. > > * Keep R fixed at 1.0 and search the [R=1,G,B] plane in device RGB space > for an [1,G,B] triple > closest to the desired chromaticity > * Do the same with fixed G=1.0, and then with fixed B=1.0 > * Use the best of the three results Yes, a 2D search would probably be desirable. > If additivity is granted, then per channel curves should basically not > influence the CIELAB gamut > hull. They will if they map 0 to !0 or 1.0 to !1.0, since they then shift the gamut boundaries ! Graeme.