Path: utzoo!attcan!uunet!zephyr.ens.tek.com!uw-beaver!mit-eddie!wuarchive!usc!ucsd!ucbvax!ucdavis!csusac!unify!unify.com!raveling
From: raveling@unify.com (Paul Raveling)
Newsgroups: comp.graphics
Subject: Re: Octree vs. Median Cut Color Quantization
Keywords: quantization, octree, median cut
Message-ID: <1990Oct26.082834@unify.com>
Date: 26 Oct 90 15:28:34 GMT
References: <579@lobster.NOSC.Mil> <1990Oct25.140850.29957@athena.mit.edu>
Sender: news@Unify.Com (news admin)
Reply-To: raveling@unify.com (Paul Raveling)
Organization: Unify Corporation, Sacramento, CA, USA
Lines: 34

In article <1990Oct25.140850.29957@athena.mit.edu>, phils@athena.mit.edu
(Philip R. Thompson) writes:
> 
> Paul Raveling has implemented such a an octree scheme for quantizing.
> You can find it in his "Img" package available on:
> 				venera.isi.edu          [~ftp/] pub
> 				expo.lcs.mit.edu        [~ftp/] contrib
> This algorithm is a little slower, uses more memory but certainly
does
> the best job of quantizing colors that I have seen, in publicly
> available quantizers.

	Actually quantization in V1.3 of the Img software set is
	a different algorithm.  It works in part by synthesizing
	a tree (not an octree); earlier versions of the Img software
	set had my earlier algorithm that did in fact operate by
	building a potentially huge octree and pruning it.

	The new algorithm generally seems to produce (IMHO) better
	image fidelity, uses less memory, and runs faster than the old
	one.  It includes color dithering, and the quantization algorithm
	tends to be better at producing a set of colors appropriate
	for a dithered image.

	If you're still using the old algorithm, it might be worth
	picking up a fresh copy of img_1.3.tar.Z and trying the new one.
	I still haven't had time to run a formal evaluation on it
	or write it up, but will try to make time for that within
	the next month or 2.


------------------
Paul Raveling
Raveling@Unify.com