Visible noise and CFA filter spectra, part two

In the preceding post, I looked at the relationship between the spectra of several cameras color filter array mediated spectral response and the noise after the camera raw files are converted to colorimetric ones.  In this post, we’ll look at some abstract responses.

The first is 2-degree CIE 1931 XYZ:

Top notch SMI, decent noise.

Now we’ll look at some Gaussian spectra. This one holds the standard deviation (sigma) to 10 nanometers, and optimizes the location of the peaks to get the highest SMI possible.

This yields impressively low noise, but an SMI that is mediocre at best.

Widening the sigma to 20 nm:

The noise is almost as low, and the SMI is quite a bit higher.

Now a very wide sigma, 50 nm.

Low SMI, higher noise.

Here are three Gaussians with both their means and sigmas optimized for accuracy:

Excellent SMI, and the noise is not bad.