That's an interesting one. On Sat, 18 Apr 2009 21:19:08 Ignacio Ruiz-de-Conejo wrote: > And I was wrong when I said number of nodes is not important, for I > think a CLUT with 65^3 nodes will reduce the space between nodes and the > bad/interpolated values, compared to a 17^3 or even a 33^3 one. > > Still, the same question apply: can anybody suggest a different approach > to build such a profile?
Actually, as I understand it, you're trying to use the ICC profile's LUT for something it hasn't been designed for. The LUT has been intended to provide support nodes for a multi-dimensional interpolation. Your values are usually unlikely to be located precisely on the nodes of the LUT, so the interpolation takes the surrounding neighbour nodes into consideration and computes a smooth transitional point location for the intermediate space. In your case, if a point is located in the "dead-man's strip" between common areas of your segmentation, then you're going to get blurring or anti-aliasing effects, that are probably not what you want (as you're looking for sharp segmentation boundaries). You're working not with a smooth profile, but with a function that is multiply non-continuous (multiply not differentiable). I guess I can see only two possibilities for this: Use a "fully" occupied LUT for your sRGB colour space (256^3 nodes) so you won't run into interpolation on 8 bit sRGB images, or use an altogether different approach (non-ICC or non- colour engine based). HTH, Guy -- Guy K. Kloss Institute of Information and Mathematical Sciences Te Kura Pūtaiao o Mōhiohio me Pāngarau Room 2.63, Quad Block A Building Massey University, Auckland, Albany Private Bag 102 904, North Shore Mail Centre voice: +64 9 414-0800 ext. 9585 fax: +64 9 441-8181 eMail: g.kl...@massey.ac.nz http://iims.massey.ac.nz ------------------------------------------------------------------------------ Stay on top of everything new and different, both inside and around Java (TM) technology - register by April 22, and save $200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco. 300 plus technical and hands-on sessions. Register today. Use priority code J9JMT32. http://p.sf.net/sfu/p _______________________________________________ Lcms-user mailing list Lcms-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lcms-user
