Ben Goren wrote:
If I'm understanding this right, since you're essentially profiling a target with (almost) monochromatic light sources, wouldn't that imply that one should be able to do the profiling with a linear RAW conversion with no white balance applied? White balance adjustment of actual pictures would then be a matter of converting to a profile with the desired white balance, no?
It's a way of measuring the spectral sensitivity of the sensors. You still have to figure out how to make a profile out of the results. One of the issues with cameras is that it comes down to a statistical argument about the "typical" spectral composition of the images that will be captured. If the sensors don't meet the Luther condition (the spectral sensitivity is different to a human observer), which is typically the case, then it's simply impossible to perfectly capture the color accurately. The best you can do is capture some range of spectral color values accurately. So in principle you optimize an RGB to XYZ transform matrix over a set of statistically "typical" spectral reflectance color values with your chosen set of "typical" illuminant spectra, to give a compromise minimum error. Measuring the camera sensors spectral sensitivities allows you to do this matrix fit over a very large number of measured or theoretical colors and illuminants. You could also arrive at several sets of matrices optimized for different situations. If you are unable to measure the sensors this way, then making a single measurement of a fixed set of colors (say a ColorChecker SG or other multi-colorant based chart) that has some statistical similarity to the real world, under a typical illuminant, is a practical alternative. In terms of white balance etc., there's nothing to stop you doing this with a matrix profile created in some other way - ie. use the matrix to convert the RGB sensor values into XYZ, and then choose a white point. Graeme Gill.
