Graeme Gill wrote: > That's one approach, although profiling using a test chart, particularly > one composed of real world spectral samples, leaves other non-linear > possibilities though. I have X-Rite "Digital Color Checker SG". They announced as real world spectrum samples chart. I've experimented a bit and found that CLUT profiles are often inadequate in darks. But matrix+shaper profile gives less color accuracy. > The basic issue is that if a sensor has spectral > sensitivities that are not a linear transform of those of a human observer, > then it simply doesn't see the world in quite the same way, and a single > matrix transform may not be the optimal way of making it approximate > a human observer. I think the transforms from linear sensor data to XYZ are always linear because law of conservation of energy and III law by Grassmann, irrespectively of human vision mismatch. I think the physically correct conversion by simple spectrum multiplication of sensor sensitivity and observer sensitivity must be done. Yes, there are the other things, but they have to be done by the next step of conversion. The linear conversion from linearised RAW to XYZ may be one universal, condition independent (ICC) profile with XYZ connection space and "absolute" output data. Yes, that profile will produce huge errors, but the next step may be other profile (may be device link), condition-dependent, which do the white balance and CLUT correction of color mismatch caused by sensors and vision difference. So I propose next workflow: 1. Condition-independent universal absolute sensor profile: 1.1 RAW data linearisation 1.2 matrix conversion to absolute XYZ 2. Condition-dependent local CLUT device-link profile with standard output, AdobeRGB1998 for example. > If something is known or assumed about the statistical distribution > of different colorant reflective spectra being photographed, then it > may be better to (effectively) use different transform matrices for > different colors. A CLUT based profile has an equivalent effect. Yes. I suppose all that can be done without condition-depended profiling "at site" have done by camera producers in camera software and proprietary RAW converters.