Recognition System for Handwritten Characters by Alternative Method using Neural Network

摘要

We propose a recognition system for handwritten characters by alternative method as a fine classification method using neural network. This method samplifies recognition problem to liner separation problem of two classes and possesses high recognition ability. When applying the new method to recognition problem of KANJI, the difficulty of large combination of categories appears. Here, we introduce dynamic learning method to cope with this problem. The dynamic learning method is able to decrease the number of modules, so that this problem can be ameliorated. We show that the proposed method can realize high recognition rate with ETL9B database of handwritten characters.