On Mon, Feb 2, 2009 at 9:31 PM, Gerhard Fuernkranz <nospam456@xxxxxx> wrote: > Pascal de Bruijn wrote: >> >> Well yes, but no software has support for such a construction, but that >> would be really cool though. >> >> I still have one remaining issue, after inspecting some of my profiles, I >> noticed the single shaper curve can be a bit bumpy. On most photo's I've >> checked so far this does not seem to have a visual consequence. > > Yes, it's my feeling as well that the shapers of "-as" or "-aS" profiles > tend to attempt to compensate the natural limitations of a matrix for > data which simply can't be modelled exactly with a matrix-based model > (for instance data from non-colorimetric color sensors, etc.). The bumps > likely improve the fit to the training set, but they don't reflect the > actual OECF and I would not be sure whether the generalization to data > not in the training set is really improved by these bumps as well. It doesn't have to be that exacting... >> However, is there a way to smooth the single shaper curve? In other words, >> not >> to change the general shape, but only to remove smaller deviations... > > You could try "-ql", this will give you smoother shapers with fewer > degrees of freedom. At least I suggest that you don't use -qu. Using a > target with more patches may possibly improve the shapers as well. I understand, however, when I use -qu instead of -qh/-qm/-ql for some targets, the curve is totally different. And when comparing to the source image the curve in the -qu generated profile seems to fit better. I'll do some more experimenting first, and possibly provide a sample this evening. Regards, Pascal de Bruijn