John Weissberg wrote: > I have noticed that the value in the wtpt tag of an Argyll-made ICC > profile varies dramatically depending on the quality level selected > via the -q and whether input/output matrices are included via the -n > tag. I would like to understand how it is that the wtpt is not > invariant with respect to these tags: When you fit a smooth regression model (smoothing splines in case of Argyll) to the measurements, then the splines will rather pass smoothly through the cloud of data points, but won't fit the data points exactly. And the "white" data point is no exception; like any other data point it may deviate more or less from the fitted curve. So for CMYK=[0,0,0,0] the splines will not predict exactly the same L*a*b* numbers as measured for the white patch(es). For relative colorimetric or perceptual intent it is however very nasty, if CMYK=[0,0,0,0] does not map exactly to paper white, as this would result in scum dots. So my understanding is that Graeme "cheats" a little bit and does not put the measured white, but the color predicted for CMYK=[0,0,0,0] by the regression model into the wtpt tag of the profile, in order to optimize the relative colorimetric and perceptual intent. In order to reduce the fitting error of the regression model at/near white, you can populate your target with more "white" patches (option "-e" of targen). Note that the dependency is O(N²), i.e. to reduce the fitting error for white by a factor of two you need to use 4 times as many white patches (this relationship will be of course distorted somewhat by the outputs tables). >> Measured whitepoint (from .ti3 file): 97.510 0.78000 -2.9600 >> Whitepoint from wtpt tag: 89.079722, 0.552168, -2.911983 This does really not look sane. Are you sure that the measurements are sane? Is the printer well-behaved (i.e. well-linearized)? But without looking at the .ti3 file it rather hard to get a clue what might be wrong.