A STUDY ON JAPANESE HISTORICAL CHARACTER RECOGNITION USING MODULAR NEURAL NETWORKS

In this research, we develop the Japanese historical character recognition system for the reading support system for Japanese historical documents. We use the directional element features as feature vectors and use the modular neural networks as pattern classification method. The modular neural networks consist of two kinds of classifiers: a rough-classifier and a set of fine-classifiers. In the rough-classifier, we use the multi-templates matching in order to find the several candidates of character categories for the input pattern. The multi-templates for each category are derived from the input samples using the Self-Organizing Maps (SOM). In the fine-classifiers, we use the multi-layered perceptions (MLP), each of which solves the two-category classification problem. The final result of character recognition is derived by selecting the MLP which has the maximum output among the set of MLPs. We also use the rough-classifier for the selection the training samples in the learning process of multi-layered perceptions in order to reduce the learning time. Through the experiments of historical character recognition for 57 character categories, we confirmed the effectiveness of our proposed method compared with the conventional research.