KANJI CHARACTER-RECOGNITION USING NEURAL NETWORKS

Japanese Kanji characters have many kinds and a complex structure compared with English and other European languages. Moreover, Kanji characters are very similar to each other. In this paper we describe multi-font Kanji character recognition using neural networks. A shift tolerant neural network architecture, a receptive field neural network architecture and a hierarchical structure were developed for Kanji character recognition. The trained neural network was tested using 12,302 characters which were selected from unknown data set and also tested using 12,379 characters from trained data set. The system achieved 99.0% accuracy for unknown data set and 99.9% accuracy for trained data set.