[mapserver-users] FW: [Algorithms] Colour reduction code

Norman Vine nhv at cape.com
Thu Jul 12 07:00:17 EDT 2001


FYI

-----Original Message-----
From: gdalgorithms-list-admin at lists.sourceforge.net
[mailto:gdalgorithms-list-admin at lists.sourceforge.net] On Behalf Of
Martin Gladnishki
Sent: Thursday, July 12, 2001 6:39 AM
To: gdalgorithms-list at lists.sourceforge.net
Subject: Re: [Algorithms] Colour reduction code


> The other thing to think about is that octree colour quantization isn't
> generally very good. I'd recommend using Wu's colour stats. from Graphics
> Gems II as a starting point. Dithering can help as well, but it depends on
> how you're using the textures. (There's a nice Siggraph 2001 paper on a
new
> error diffusion dither that's worth trying.

I agree, octrees are no good for quantization. The best color quantization
algorithm I've ever seen is the quantization with Cohonen neural network.
Gives _much_, _much_ better results than octrees.
But you still have to deal with the alpha's anyway :)

Give it a try:
http://members.ozemail.com.au/~dekker/NEUQUANT.HTML

Cheers,
Martin


_______________________________________________
GDAlgorithms-list mailing list
GDAlgorithms-list at lists.sourceforge.net
http://lists.sourceforge.net/lists/listinfo/gdalgorithms-list




More information about the mapserver-users mailing list