Any pointers to algorithms for reducing RGB888 images to 8-bit + clut?
The generic terms to google for are "color quantization" and "error
diffusion dithering". The first is the problem of finding a (small)
colour palette that covers the color gamut of some given image(s). The
latter is about using pixels of the limited palette to represent the
original true colour image as good as possible.
Depending on how much time you have at your hands, and depending on how
much cycles you can spend, there are several algorithms to choose from. In
particular, there are other colour dithering algorithms than error
diffusion, such as halftoning or other ordered dithers or even
sophisticated iterative solutions based on energy minimization.
Quantization is also a wider field, colour quantization is a special case
of "vector quantization". On the other hand, human perception of colours
plays a role here. So if you want to go for the best possible color
reduced images, you might want to do all computation in a colour space
that models perceived contrast of colours.
Or, if that is an option, be inspired by open source programs like
ppmquant (of the netpbm package).
And as Ian said, don't do colour reduction when you really want to
compress images. Compression algorithms can make much smarter choices
about which redundancy to get rid of.
