Image color quantization procedure (FAD)

El_Choni · Post by **El_Choni** » Sun Apr 11, 2004 7:42 am

The 24 bit values are in first instance reduced to 15 bit for that purpose. I use 5-5-5 for RGB, but you can use 6-4-5 if that suits you best, or any other combination. Sorry for the badly commented code. I'll try to clean it a bit when I return to Spain.

LCD · Post by **LCD** » Sun Apr 11, 2004 5:49 pm

El_Choni wrote:The 24 bit values are in first instance reduced to 15 bit for that purpose. I use 5-5-5 for RGB, but you can use 6-4-5 if that suits you best, or any other combination. Sorry for the badly commented code. I'll try to clean it a bit when I return to Spain.

I have two purposes for quantisation: For the Spec256 Emulator (256 colors out of 24 Bit palette), where I can use the faked octree, that is damn slow but offers excellent quality of choosen palette.
The second quantisation is to SAM Coupe Palette (64 colours in 2 bright levels=128 colors, 16 on screen without using line interupt), which works very good. I also have the possibility to choose the percentage of error distribution, because the results can be improved significantly for less colours if I choose for example 70-80% of the error for distribution.

Oh yes, don't worry about the comments too much, I just cannot speak very good spain language

. And I saw C++ source code that got exactly not even a single comment.