Xref: utzoo comp.graphics:5851 comp.ivideodisc:209
Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!rutgers!cs.utexas.edu!uunet!seismo!esosun!cogen!celerity!celit!billd
From: billd@celerity.UUCP (Bill Davidson)
Newsgroups: comp.graphics,comp.ivideodisc
Subject: more color quantization
Message-ID: <313@celit.UUCP>
Date: 24 May 89 08:33:54 GMT
References: <12700@shamash.cdc.com> <30@oink.UUCP>
Sender: news@celerity
Reply-To: billd@celerity (Bill Davidson)
Organization: FPS Computing Inc., San Diego CA
Lines: 17

Well, I tried my idea for modifying the median cut method by cutting
at a point in which each half of the array represented roughly the
same number of pixels in the original image.  I got varying results.
Some images were *slightly* degraded.  You could only tell in detailed
spots with rare colors.  This is rare in the images I have.  Several
images showed no discernable affect over just using a standard median
cut and doing a weighted average for each box.  A few images were
vastly improved.  Ray-tracing images which involved shiny reflective
spheres lost almost all banding!  I really had to look close to see
any at all.  The background had been badly banded before and it's banding
disappeared.  I expected a slight improvement and I got a vast
improvement.  So, what's the verdict?  Did I re-invent the wheel
(or at least a spoke :-) or what?  If somebody doesn't say something
first, I'll probably try to figure the mean quantization error for
the entire image with of these variants and try to figure best/worst
case scenarios.
Bill Davidson		...!{ucsd,sdsu,fpssun,cogen,chip,photon}!celerity!billd