ニューラルネットワークを用いた類似文字のリジェクト処理

概要

論文の詳細を見る
This paper reports a recognition system for similar handwritten Chinese characters. It uses a neural network. Introduction of a novel rejection method into the system is expected to enhance recognition rates. The rejection method removes characters that pose a high probability of incorrect judgment; these characters are sent to another recognition system (relief system) for correct judgment. We propose this new rejection method which uses final output values in the neural network and show experimental results which confirm its advantages. The recognition system including the relief system achieves much higher recognition rates than without a relief system.