Yesterday I published a post that took a set of Sigma Foveon silicon spectral responses and a hot mirror spectral transmission spectrum, calculated an optimal compromise matrix, and evaluated the color accuracy of the camera using the Macbeth CC24 patch set for both training and evaluation.
A reader questioned the accuracy of the curves supplied to me. Fair enough, although the curves looked reasonable to me.
Iliah Borg, whom I greatly respect, send me a PDF of a paper from the Journal of the Optical Society of America (Maryam Mohammadzadeh Darrodi, Graham Finlayson, Teresa Goodman, and Michal Mackiewicz, Reference data set for camera spectral sensitivity estimation, JOSA Volume 32, Number 3, March 2015, pp 381-391), and an Excel spreadsheet with the curves.
Here is the result with Simplex optimization:
Quite similar to the Spectral Metamerism Index I got with Ted’s data. 82.2, vs the 82 I got before.
Here are the results using the genetic algorithm with a population of half a million.
The GA did a bit better than Simplex.
The GA converged fairly rapidly:
The GA eats a lot of computer cycles: