Hi Magne, Sorry about the delay, I have little free time lately and I'm checking the lcms mailing list once a week and even less sometimes...
Ok, so now for your question. Yes, you are right Lab to whatever, as long as you begin with Lab, takes longer than some other transforms. The reason is tricky: basically Lab as input needs a additional interpolation across three tone curves. Why? Well, that's the tricky part. All precalculated transforms uses 3D (or N-D) tables with trilinear or tetrahedral interpolation. The input space is used to index the cube and each grid node of cube is populated with values of output space. This is very convenient for spaces like, for example, RGB, where each component goes from (in 16 bits) zero to 0xFFFF. Then 0 indexes the first node and 0xFFFF the last node. All middle nodes are placed in intermediate locations. Then comes Lab. ICC did choose a weird encoding for V2 Lab, so L* for example, which goes from 0 to 100, is encoded 0x0000 for zero and 0xFF00 for 100. Please note with this encoding 0xFF00 does NOT index the last node on the grid (that would be 0xFFFF). lcms needs to do something to place white over a node. That is important to avoid "scum dot" and other artifacts. So, in order to map 0...0xFF00 to 0..0xFFFF range, lcms uses a tone curve with 257 entries. Obviously that doesn't apply on RGB to RGB or even on CMYK to CMYK transforms, only when the indexing space is Lab (as is your case) you have this penalty in order to increase accuracy. Next versions of lcms would use V4 ICC encoding, which solves partially the problem, but I suspect that would not apply in your case since you need V2 Lab encoding for doing your calculations. Hope this helps All the best Marti Maria The littleCMS project www.littlecms.com -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Magne Nilsen/ETC Sent: Friday, February 29, 2008 2:38 AM To: lcms-user@lists.sourceforge.net Subject: [Lcms-user] Speed of RGB -> Lab ->RGB Hi, I have data converted into 16 bit ProPhoto RGB. I want to as fast as possible convert this to Lab, do some changes in Lab mode and convert back to ProPhoto RGB again. The first conversion is fine: ------------------------------------------------------- //Convert ProPhotoRGB to ProPhotoLAB lcmsPPProfile := cmsOpenProfileFromMem(rStreamProPhoto.Memory,rStreamProPhoto.Size); lcmsLABProfile := cmsCreateLabProfile(NIL); lcmsTransform := cmsCreateTransform( lcmsPPProfile, TYPE_RGB_16, lcmsLABProfile, TYPE_Lab_16, INTENT_PERCEPTUAL, 0 ); //normal precalc for y := 0 to CurrentBitmap.Height-1 do cmsDoTransform(lcmsTransform, CurrentBitmap.ScanLine[y], CurrentBitmap.ScanLine[y], CurrentBitmap.Width); cmsDeleteTransform(lcmsTransform); //Timing inc createTransform on 6MP image //1340 ms w/ 0 in precalc!! ------------------------------------------------------- and 1340 ms is much faster than I am able to do this... and if I have the transform cached, the time goes down a little more for the Transform only as well. The LAB image is verified to be exactly as expected. BUT, even when doing nothing to the image, I have problems getting this equally fast back to ProPhoto RGB: ------------------------------------------------------- lcmsTransform := cmsCreateTransform( lcmsLABProfile, TYPE_Lab_16, lcmsPPProfile, TYPE_RGB_16, INTENT_PERCEPTUAL, 0 ); //normal precalc for y := 0 to CurrentBitmap.Height-1 do cmsDoTransform(lcmsTransform, CurrentBitmap.ScanLine[y], CurrentBitmap.ScanLine[y], CurrentBitmap.Width); cmsDeleteTransform(lcmsTransform); //Timing inc createTransform on 6MP image //3230 ms w/ 0 in precalc!! ------------------------------------------------------- Why does that 2nd transform take that much more time? 3230 ms vs. 1340 ms ? Is this just the nature of things, or am I doing something wrong? Or - is there any alternative ways of doing this... ? Any hints would be appreciated... regards Magne ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Lcms-user mailing list Lcms-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lcms-user ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Lcms-user mailing list Lcms-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lcms-user
