Edmund,

err, I have googled both many times, but that doesn't turn me into a colour scientist.

Best regards

Erik

Erik,

Yeah, the standard colorimetric x, y and z are rgb observer functions which are obtained by passing light through a (spectral) filter which basically corresponds (after various neural encoding issues) to the cone pigments in the average eye - 3 types of cones.

So some illuminant spectrum bounces off some object, into your 'standard observer' eye, gets filtered through say the "r" cone pigment, then the power of the remaining light gets integrated. Hey now you have 'x'.

A camera needs to somehow recreate these integrals if it wants to recreate the observer perceptions. With 3 filters that means having filters which need to basically be independent linear combinations of the cone pigment spectra, I guess. This is called the Luther-Ives condition.

A camera which does not satisfy Luther Ives will suffer from so-called metameric failure, ie. it will differentiate colors seen as identical by a human observer, or fail to differentiate colors perceived as different. Of course no camera satisifies the Luther Ives condition, but high-end colorimeters try to, and so do the computations in spectroradiometers. I believe a standard research technique is to aim a teleradiometer at a scene and a camera and compare the two. Such experiments where discussed at ICC meetings when talking about formats for camera data.

Anyway my comment to your comment

is that in my opinion with a 3 filter CFA array the ideal "slopes" of all the CFA spectral functions satisfying Luther-Ives here are in some way known - there is little tolerance for variation because they are determined by the cone functions which are physiological - if the standard observer were to donate his body to science, we could cut his eye open, extract some cones and measure the pigment transmisittivity.

What happens a lot with real cameras -I believe- is that the x'y'z' in the camera are not "orthogonal" enough, and then the determination of the xyz becomes imprecise. I don't really understand this part, but I see it happen a lot, and I've talked to people like the ex-CEO of Hasselblad who told me this is the problem they faced *in practice*. Or rather, he told me this is the problem his competitors with dLSRs faced in practice

My feeling is that the A7IIR is facing this type of issue. Or quite possibly I'm wrong.

Edmund