Thread: [Lcms-user] Creating a display profile using lcms2
An ICC-based CMM for color management
Brought to you by:
mm2
From: Richard H. <hug...@gm...> - 2012-12-03 17:13:14
|
I'm trying to validate the results of argyllcms-generated profiles in my effort to test the linux color managed stack we have. It's part of my hidden agenda to get CM working well and hopefully officially supported in RHEL. Is there any easy way for lcms2 to build me a matrix or LUT profile given a a ti3 file? It doesn't have to be v2 or v4, I just need something that will give similar results to the profile created by colprof. A sanity check if you like. Any pointers very welcome. Thanks, Richard. |
From: Graeme G. <gr...@ar...> - 2012-12-03 23:22:58
|
Richard Hughes wrote: > Is there any easy way for lcms2 to build me a matrix or LUT profile > given a a ti3 file? It doesn't have to be v2 or v4, I just need > something that will give similar results to the profile created by > colprof. A sanity check if you like. Any pointers very welcome. You probably want to check out the ICC SampleICC project if you want alternate (free) profiler functionality. I don't actually know if it provides the functionality you are after though.. Graeme Gill. |
From: Gerhard F. <nos...@gm...> - 2012-12-04 10:01:22
|
Am 03.12.2012 18:13, schrieb Richard Hughes: > I'm trying to validate the results of argyllcms-generated profiles in > my effort to test the linux color managed stack we have. It's part of > my hidden agenda to get CM working well and hopefully officially > supported in RHEL. > > Is there any easy way for lcms2 to build me a matrix or LUT profile > given a a ti3 file? It doesn't have to be v2 or v4, I just need > something that will give similar results to the profile created by > colprof. A sanity check if you like. Any pointers very welcome. Richard, might cross validation be an option for you? For instance, N-fold CV: Split the color samples into say N=10 subsets, build a profile from the union of the subsets #2,...,#N and verify this profile against the samples from subset #1, then build another profile from subsets #1,#3,...,#N and verify against subset #2, and so on, until all N profiles have been built and verified. Particulary for models with a larger number of parameters (e.g. LUT profiles), CV usually provides a more meaningful goodness of fit indication anyway than the residuals of the training set. Best Regards, Gerhard |