On Jan 17, 2013, at 10:43 AM, Gerhard Fuernkranz <nospam456@xxxxxx> wrote: > Am 17.01.2013 18:10, schrieb שחר קלינגר: > >> I see. I think I understand what's going on. Does this mean that because you >> cannot measure the scanner's sensors' response directly, you're using a >> printed reference to do that, therefore you can't really know the scanner's >> RGB primaries (or it doesn't even matter)? > > An input device does not have any "primaries" in the sense of an additive > output device. The red/gree/blue matrix colums tags in the profile are just > what their tag names sugget, namly the colums of a transformation matrix, > which is part of a matrix/TRC model. And the model parameters (including the > matrix) are derived by minimizing the "overall error" for all the RGB/XYZ > pairs given in the .ti3 file. Another way to look at it that might help: you can expose the input device to light of any spectral distribution you can physically generate, and there will be a corresponding RGB value. It might not be a very useful RGB value -- it might not be very bright, or it might be noisy, or it might not be significantly different from the value generated from some quite different spectral distribution, or it might suffer from any other failing you can imagine. But feed a spectrum to an input device and you'll get a predictable RGB value out of it. The ultimate job of the input profile is to try to figure out what real-world spectrum produced a given RGB value -- with the caveat that, rather than actual spectral distributions, XYZ (or some other device-independent color space) values are used as a ``good enough'' proxy for spectra. Since the concept of "primaries" only applies to certain ways of generating spectral distributions, and since an input device may well be exposed to spectral distributions that were generated by completely different means than a mixture of a few (or even a great many) "primaries," the concept simply doesn't apply to an input profile. Cheers, b&