gwoho@nntp-server.caltech.edu (g liu) (12/05/90)
what is a good way to chose a palette for a picture? when picking the palette, the number of bits per primary is reduced, and the number of colors is llimited. the algorithm must produce, when possable, colors which suround the actual colors in the picture. i.e. if 5% of pixels are, say 0, 5% are 100, and 90% are 50, and the program is told to pick 2 colors, it should pick 0 and 100, even though most pixels are 50. this is because 50 can be made by dithering the 0 and 100. if it chose, say 0 and 50, then the f-s error terms when mapping the picture would diverge, and the picture would look terible. i want the algorithm that produces the best results; i dont care about cpu time, as long as it doesn't take a few days. gwoho liu.
gwoho@nntp-server.caltech.edu (g liu) (12/06/90)
i realize that there is an refrenence to this in the frequently asked question. the algorithm there is crummy. pbmplus if loaded with bugs, and its quantization algorithm doesn't work well if the color resolution is lowered a lot (i.e. going from 8 bits per primary to 2 bits per primary). also it doesn't do the first pixel of every other line, does something strange at the end of everyother line, and it the arithmatic overflows in several places if i use 16bit ints, 32bit longs. i want a diferent algorithm: one that handles lowering of color resolution and limiting of the palette better. gwoho liu.