Originally posted by Class A If the metamerism index is defined that way -- assuming a standard human observer (surely individuals have differing colour perception) -- than that'll be fine.
I wasn't sure whether, e.g., DxOMark just measures RGB separation.
I could image that different sensors use different colour filters to achieve different image aesthetics. Surely a lot of things (colour separation, colour fidelity, sensitivity) must be balanced and there are more than just one "correct" solutions.
I might be wrong, though.
Well, Unsinkable II thinks I know nothing about this, so take my response with a grain af salt
Neverthless, I'd say that the usage of different colour filters to achieve different image aesthetics is not an option. Humans perceive colors they way
they do. Their perception was measured in field tests and turned into norm ISO 17321.
Actually, the studies for these norms were originally done to classify the naturality of light sources but serve well to test camera colors as well.
The test basically measures the deviation of a camera's RGB response from a standardized RGB response, for something like twenty or so color patches produced in a standardized process (not cheap to aquire). You see the patches in DxO's full color spectrum charts. The standardized RGB response is derived from a patch with known tristimulus values where a majority of humans say to see the same color.
In order to achieve a metamerism index of 100, the color filters would have to replicate the exact spectral response curve of the three types of human cones. This article contains interesting information about the spectral sensitivity of the eye and semiconductors:
http://www.astro.rug.nl/~sctrager/teaching/observing_techniques/Detectors1.pdf
It is known that Canon has a not very selective red channel (and DxO measurements confirm this, cf. the RGB calibration matrix coefficients). Therefore, the color noise is a bit amplified and color sensitivity is a bit reduced. However and surprsingly, the metamerism index for Canon cameras is not worse. The metamerism index doesn't change if you add one channel response curve to another if they originally didn't overlap.