Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!ucla-cs!sdcrdcf!trwrb!aero!venera.isi.edu!lmiller
From: lmiller@venera.isi.edu (Larry Miller)
Newsgroups: comp.graphics
Subject: Re: Preview devices
Message-ID: <2554@venera.isi.edu>
Date: Thu, 23-Apr-87 11:17:31 EDT
Article-I.D.: venera.2554
Posted: Thu Apr 23 11:17:31 1987
Date-Received: Sun, 26-Apr-87 19:39:10 EDT
References: <3017@sdcsvax.UCSD.EDU> <1709@cbmvax.cbmvax.cbm.UUCP>
Reply-To: lmiller@venera.isi.edu.UUCP (Larry Miller)
Organization: Information Sciences Institute, Univ. of So. California
Lines: 42

In article <1709@cbmvax.cbmvax.cbm.UUCP> hedley@cbmvax.UUCP (Hedley Davis) writes:
><munch>
>
>As a breif aside, you should look at the '83 SIGGRAPH proceedings which
>contains a ( now classic ) article describing color compression in
>images. The article deals with the 'Median Cut' algolrithm for color
>table entry allocation. Very impressive two bit per pixel color images
>are presented. 
>
>If you use the techniques in this article, and the hold and modify mode
>of the amiga, you can acheive surprising results. Sure this is compute
>intensive, but it can give you some of the best results I've ever seen
>on an Amiga. 
>
>Hedley 

Here is a citation (in refer format) to another article by the same author,
describing the technique.  I've implemented this on 24 bit/pixel digitized
maps on an IRIS, reduced to 4 bits per pixel.  The results, however, were
not too good.  It was necessary to change the algorithm somewhat to allow
for very low frequency shades being retained in the image (lat/long lines,
for example).

Larry Miller
lmiller@venera.isi.edu

--------------------------- CUT HERE --------------------------------- 
%A Paul Heckbert
%T Color Image Quantization for Frame Buffer Display
%J Computer Graphics
%D July, 1982
%V 16
%N 3
%P 297-307
%K Graphics, Dither, Color images
%X Presents a method for selecting the N best colors to represent an image
digitized at M (M >> N) colors.  The methods presented are ``uniform
quantization'' and ``median cut.''  In uniform quantization, the N most
frequent colors are selected.  Median cut uses an adaptvie algorithm to
find the N ``best'' colors, and generally produces better results.  Typical
quantization levels are eight bits per pixel.  Dithering also increases the
quality of the final images.