Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!math.ohio-state.edu!caen!msuinfo!harbinger.cc.monash.edu.au!news.cs.su.oz.au!metro!news
From: "" <alison@atp.biochem.usyd.edu.au>
Subject: Question: Pattern recognition problem
Message-ID: <16841.alison@atp.biochem.usyd.edu.au>
X-Minuet-Version: Minuet1.0_Beta_8
Sender: news@ucc.su.OZ.AU
Nntp-Posting-Host: lennon.biochem.usyd.edu.au
X-Popmail-Charset: English
Reply-To: <ajl@biochem.usyd.edu.au>
Organization: Information Services, Sydney University, Sydney, NSW, Australia
Date: Thu, 3 Nov 1994 19:04:10 GMT
Lines: 28

I am considering using a fully-connected feedforward neural network trained
using backpropagation for pattern recognition.  I will have a library of many
(eventually > 100) stock images and I want to determine whether a new sample
matches a specific one of these images.

I was contemplating training a different network for each of the stock images
using frame-shifted, slightly rotated (1-2 degrees maximum) and noise-added
variations of all the stock images, and having a single output unit which would
be set to 1 if the sample represented the desired stock image, and 0 if not.
The weights of the trained networks would be stored in a file to be accessed by
the runtime system.  The images will be in the format of 100x100 bitmaps.

Has anyone had any experience of this type of problem?  If so, have you any
comments or suggestions which might help.  Should I, for example, be looking at
other network architectures and/or training methods; compressing the bitmaps,
etc.

Thanks.

Please reply by email to 100240,2017@compuserve.com as I do not have regular
access to this bulletin board.  I ll summarise to the group.
----------------------------------------------------------------------
Alison Lennon                         email : ajl@biochem.usyd.edu.au            
Department of Biochemistry,          
University of Sydney,                 Telephone : +61 2 351 3746
NSW, 2006,                            Fax       : +61 2 351 4726 
Australia.
----------------------------------------------------------------------
