In my experience, which of course is very limited as I don't work for a camera or sensor company, the simple matrix method is ok if the matrix is derived from a test image in the lighting which is then used for the subject matter. I have used commercial software to create usable general-purpose camera profiles, but in difficult or mixed light a simple matrix derived from actual measurement of a test image in that light will always win out. To my eyes anyway. At the level of the ICC, I have been lobbying for the inclusion of fields in TIFF/EP for a spectral power measurement of the illuminant of a scene, and for inclusion of spectral sensitivity curves for the cells composing the sensor. There are now -medium expensive- solutions for field measurement of both these curves, and the information gathered could be processed by any Raw converter to derive decent color. I have also lobbied to request that all Raw converters be capable of picking up manufacturer included data including matrices and rendering profiles when present, but there is industry resistance as the camera manufacturers don't want to publicize either sensor data or matrices and renderings, and Adobe likes to use only their own data. Edmund On Sun, Jul 27, 2008 at 1:23 PM, Gerhard Fuernkranz <nospam456@xxxxxx> wrote: > Klaus Karcher wrote: >> edmund ronald wrote: >>> In my experience, current dSLRs are pretty well linearized in Raw mode. >>> Obtaining decent color from them is a matter of locating the primaries. >>> The raw2xyz transform can be considered to be a matrix. >>> The problem resolves into a one line Matlab or Ocatve program. >>> Whether you want to deal with flare, or other secondary phenomena, is >>> a different issue. >> >> Is it usual to get primaries way beyond the spectral locus or even >> with negative components (e.g. like those in Milan's profiles) with >> current cameras? > > It depends on the method being used. There are several methods proposed > in the literature for deriving the 3x3 matrix, which claim to work > better than a simple minimization of an error metric in a least squares > sense, for a given a calibration target. Some of them seem to make use > of additional constraints like exact mapping of white, positivity of the > captured spectra, or particular assumptions about the correlation of the > captured spectra (e.g. smoothness of the spectra), etc. I haven't yet > looked into these subjects in detail though. > > Regards, > Gerhard > > >