I'm in fact wondering whether it might be worth to attempt to fit a very smooth RGB/CMYK -> variance RSPL mapping to this data, and to use this RSPL during profile generation to obtain an individual weight for each training set data point (i.e. weight[i]=1/variance[i], probably normalized such that the sum of the weights is 1.0 in order that the per data point weights do not interfere to much with the existing avgdev -> raw smoothness mapping).
that's what I was thinking just now; without all the math, of course :) but the basic idea: - you either repeat patches in the target (a new targen option), making it 8,000 instead of 1,000, or just measure 8 prints of the same 1,000 patch target. - a new column is added to the ti3, "patch reliability" or something like that - you can save a ti1 with the added colum, for specific future use in same device kind+ink+paper in which you wouldn't have to measure 8 times, just once. - when building a profile, this info would be taken into account so that the fitting error tolerance in each point would be proportional to its repeatability, thus allowing for an improvement in smoothness by relaxing tolerances in patches with high variation.
sounds very good!! a way to achieve smoothness without sacrificing precision where it matters. by the way, the fitting error tolerance should be different in different parts of the Lab space... like it should be in dECMC instead of dE, or just take the tolerance calculated from the point reliability and multiply it by a factor dECMC/dE for that part of space... of course, switchable parameters because we could otherwise end up with too relaxed tolerances!
-- Roberto Michelena Infinitek Lima, Peru