On Jul 4, 2013, at 12:56 AM, Torsten Bronger <bronger@xxxxxxxxxxxxxxxxxxxxx> wrote: > As far as I can see, even if you take into account that all other > patches have to be fitted, too, this (purely mathematically) results > in a matrix A * M, where A has only diagonal elements. The math says that should be the case with ideal devices, yes. But, though cameras are *close* to being ideal linear devices, in reality, they aren't *actually* ideal linear devices. And errors in white balance will only serve to multiply the nonlinearity. Imagine an extreme example, such as profiling not the output of dcraw but the output of a typical consumer raw development engine, such as Adobe Camera Raw or Canon's Digital Photo Professional. Either development is going to have not just white balance, but an S-curve and wacky color matrices and what-not all baked in. And then there's gamma to boot. I'm not even sure that the total transformation is theoretically reversible or invertible, and profiling software doesn't pretend to try. Even without that type of munging, the camera deviates from perfect linearity in exactly the same type of chaotic manner, but just to a *much* lesser degree. Think about it: if the camera actually *was* perfectly linear, especially if it also met Luther-Ives, there'd be no need for profiling. You would just tag the file with a simple profile in the style of a working space with the camera's primaries and call it a day -- and, in this perfect world, for a given camera model, the primaries would be the same to the Nth digit for all cameras. In reality, the goal of profiling is to characterize how far the camera deviates from ideal. And, to do that effectively, you want to eliminate, or at least minimize, all other influences. Cheers, b&