Xref: utzoo comp.graphics:2349 comp.sources.wanted:3953
Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!purdue!decwrl!decvax!mandrill!hal!uccba!ucqais!finegan
From: finegan@ucqais.uc.edu (Mike Finegan)
Newsgroups: comp.graphics,comp.sources.wanted
Subject: Re: Clustering Algorithms
Message-ID: <1419@ucqais.uc.edu>
Date: 25 Apr 88 06:00:23 GMT
References: <615@ubu.warwick.UUCP> <6196@cit-vax.Caltech.Edu> <655@esl.UUCP>
Distribution: comp.graphics,comp.sources.wanted
Organization: Univ of Cincinnati, College of Business Admin.
Lines: 20
Keywords: clustering, ISODATA
Summary: ISODATA

In article <655@esl.UUCP>, rww@esl.UUCP (Richard W. Webb) writes:
> Hello,
> 
> 	I have heard of a program called ISODATA that does some clustering,
>     but I don't know how to get it or what really does.
> 
I used ISODATA to cluster 3-D vectors (RGB, HSI, etc.) for image processing
segmentation. The authors are Ball & Hall - Stanford Research Labs (sic?).
The algorithm is presented fairly completely in the report (dated 1965). The
parameters which control the clustering are cryptic (i.e. heuristic ...) and
it uses large amounts of memory. It worked okay for my application, but I
didn't have the patience to try every combination of process parameters, to
see if I was getting optimum clustering or not. 
    You would probably be better off with another technique, unless you have
very few data points (or not too many dimensions). If you really want code -
I could give you what I have - it was written for a class and could be cleaned 
up. Maybe you could send me some other clustering algorithms in return ? I
have a copy of the report - its about 100 pages - but the algorithm is about 
10 or 20, with flow charts, diagrams, formulae, etc.
				Mike Finegan   ...uccba!finegan