Bart, the top layer is blue-weighted. The second layer is much closer to luminance. Thus, without assumptions about the way spatial frequencies of the luminance and chromaticity of the original scene vary, is is not possible to "develop" a colorimetric RGB image with implicit luminance at the resolution of the top layer.
Hi Jim,
I agree that the top layer doesn't strictly record Luminance, that's why I said "The 'top layer' of the Foveon Quattro sensor captures
mostly a monochrome image,
good for luminance resolution." Good is to be understood as useful, but not accurate. I know it is relatively Blue weighted in sensitivity, but Blue contributes very little to the image's Luminance component (Blue is typically weighted as contributing 7.2% to total Luminance).
So it still makes a nice, not accurate (because overweighted in Blue and Red) but nice, substitute for Luminance from which a useful contribution to the other channels can be calculated. Blue wavelengths are also relatively the least affected by diffraction, and the top layer exhibits the least diffusion/scatter.
Cliff Rames made a
nice simulation that shows how the top channel can be (ab)used to copy its high spatial frequency signal to the other layers, resulting in only a bit of inaccuracy at sharp edge transitions but visually still acceptable.
The top layer has its peak at a wavelength that plays almost no part in calculating luminance from spectra.
In all of the above, I am defining luminance as the Y component of 1931 CIE XYZ.
Correct, colorimetrically it is not the same, but then again it's still useful enough. As Clif said, it "might work because it is known that high-frequencies in natural images are highly correlated among the RGB channels."
Cheers,
Bart