Hi Gerhard, I'm taking the freedom to CC to mailing list, I think this could be of interest.
>Please also look at my attached images. >Both are created in the following way: I'm not attaching the images, but the effect Gerhard is describing is a big amount of noise, specially on dark areas. Matrix-shaper noise is noticeably lower that CLUT based. >The 1st image was transformed with a matrix/shaper profile I created >myself (lower quality, agv dE = 2.6, max dE = 12) and the 2nd image was >converted with a profile genereated by your new LCMS profiler. Both were >sharpened with the same amount. >Similar to Jack's pictures, my image, if converted with the LCMS >profile, shows more noise artifacts in the dark areas than a conversion >with my matrix-shaper profile. So appearently "high-quality" profiles >tend to boost the noise which is present in the image. I have also made >some experiments with various profiles I generated myself. With up to 2 >polynomial orders (10 terms) the result are nearly as good as with my >matrix profile, with >= 3 polynomial orders (>= 20 terms) my profiles >show very similar artifacts, and with 50 terms my profiles are much >worse than yours (with regard to these kind of artifacts), although the >avg/max dE numbers are pretty good. >I'm just wondering whether these kind of artifacts are really a >necessary property of "good profiles" or whether they would be >avoidable? I also still don't yet understand why the "good profiles" >boost the noise more than a matrix/shaper profile. Any idea? The true reason is already unknown to me, but here are some thoughts A profile, for "good" operation needs to be smooth. This is done always in matrix shaper because the nature of implied math model. In such profiles you can check the patches used in training and other patches not used and the resulting dE tends to remain same. In the other hand, CLUT based can exhibit discontinuities. Thus, a CLUT based profile can model the weird behavior some devices have. But at a price. One should be careful on these profiles, because if they are not smooth, they can fit almost perfectly patches used in training, but be very bad on patches not used. For example, a profile with dE of about 0.5 can give dE > 10 on new patches. These profiles are unusable, despite the good dE. On more terms in regression, more forced is the gamut to fit outliers, and less smooth is whole profile. So, using 50 terms is good as far as 20 or less terms gives approximately same result. Another question is the hard non perceptually uniform space used. In last revision of profiler I forced to use gamma of 2.2 and more. This was not only because the 8 bits stuff, but because gamma 1 has many other additional difficulties. Noise is one of them. Gamma 1.0 means almost all values of 0..ffff are used to encode light zone. Dark part gets squeezed into relatively few codes. And what happens if these critical codes are holding noise? This is another reason because I always recommend to avoid gamma 1.0 and use 2.2...2-4 if possible. Regards, Marti Maria. ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Lcms-user mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/lcms-user
