Summary
International Symposium on Nonlinear Theory and its Applications
2009
Session Number:B4L-D
Session:
Number:B4L-D4
Rejection Judgment Method Using Neural Networks for Character Recognition
Yoshimasa Kimura, Hiroyuki Nishi, Eiichi Mukai,
pp.-
Publication Date:2009/10/18
Online ISSN:2188-5079
DOI:10.34385/proc.43.B4L-D4
PDF download (201.1KB)
Summary:
We propose a novel rejection judgment method using neural networks for character recognition. The neural networks take discriminant measures of an input pattern as input and learn so as to distinguish correct recognized and misrecognized patterns. This forms a nonlinear decision boundary which separates both patterns in a discriminant measure space, and enables the rejection of misrecognized patterns. Neural networks achieved higher performance for rejection judgment on similar shaped Kanji pattern recognition than the traditional method. Moreover, superior performance was obtained by transforming the discriminant measures in the form of the logarithm.