Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!sol.ctr.columbia.edu!cica!iuvax!rutgers!deejay!otello!idea!argo!renzo
From: renzo@argo.sublink.org (Renzo Tomaselli)
Newsgroups: comp.ai.neural-nets
Subject: pattern recognition and NN
Keywords: pattern nn ocr
Message-ID: <676@argo.sublink.org>
Date: 26 Sep 90 10:16:09 GMT
Organization: TecnoTP snc  Pergine (TN - Italy)
Lines: 20


I'm interested in pattern recognition of typewritten characters from scanned
monochrome images. Typical decision-tree methods fail where lack of geometrical
connection exists, usually because of poor image quality. 
I'm totally new to NN; however it seems to provide an interesting alternative
to conventional approaches. The case of nets transforming an input pattern to
a multidimensional point in a feature space is full of promises.
I would appreciate any help or suggestion in moving these rough ideas into
some more realistic approach. Might any NN simulator be useful ? I'm rather
confused about the kind of NN paradigm more suitable for the purpose.
Speed wouldn't be a concern at this point; I'm approaching the problem from
a theoretical point of view.
Thanks in advance to anyone who will provide me some help.

						Renzo Tomaselli

-----------------------------------------------------------------------------
TecnoTP s.n.c.    Via Paludi, 52  I-38057 Pergine Valsugana - Trento ITALY
Tel./Fax ++39 461 533629       renzo@argo.sublink.ORG
-----------------------------------------------------------------------------