Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!swiss.ans.net!newsgate.watson.ibm.com!hawnews.watson.ibm.com!hawnews!admin!danebury!ricky
From: ricky@hursley.ibm.com (Rick Turner)
Subject: Re: Multispectral classification
Originator: news@hercules.hursley.ibm.com
Message-ID: <1994Nov28.083401.31044@hursley.ibm.com>
Sender: news@hursley.ibm.com (Usenet news handler)
Date: Mon, 28 Nov 1994 08:34:01 GMT
Reply-To: shadowfax@vnet.ibm.com
Disclaimer: This posting represents the poster's views, not necessarily those of IBM.
References:  <Pine.SUN.3.91.941124150534.7171G-100000@fellini>
Nntp-Posting-Host: danebury.hursley.ibm.com
Organization: IBM UK Laboratories Ltd.
Lines: 12

If you want unsupervised classification, then no, there are few things
better than the ISODATA algorithm. Probably, what you ought to do is to
switch to a supervised method and use a Bayesian classifier. If you do 
this, then you can include things like texture, slope and aspect, or
whatever. They are simply extra image 'bands' fed into the algorithm.

I have done a fair bit of work in this area classifying SAR and mixed
SAR/optical image sets, and this method works quite well.

Drop me an e-mail offline if you want more info.

Rick
