I implemented something like this for an application I wrote once. The algorithm you're implementing chooses N evenly-spaced 3D grid points in the RGB color space. I found that this didn't give a terribly good result, because our eyes are far from being equally sensitive to changes in red, green, and blue at different points in the color space.

Anyone know of a better algorithm? As a first cut, probably using HSV space would be better, but you wouldn't want to sample it evenly -- you'd probably use a whole bunch of gradations in hue, but only a few in saturation and value.

I probably just ought to spend some quality time with Google. Given how important this is for things like MPEG encoding, I'm sure somebody's come up with a color space that approximates an even spreading of perceptual differences. Well, "approximates" it for "most" people, or something.

Then you'd just have to implement a "red/green colorblind" mode that a user could select, with a very different sampling space...