A Study on Japanese Historical Character Recognition Using Modular Neural Networks

It is fundamental work to translate the historical characters called "kuzushi-ji" into the contemporary characters in Japanese historical studies. In this paper, we develop the Japanese historical character recognition system using the directional element features and modular neural networks. Modular neural networks consist of two kinds of classifiers: a rough classifier to find the several candidates of categories for the input pattern, and a set of fine-classifiers that determine the category of the input pattern as the final result of character recognition. We construct the rough-classifier using the self-organizing maps (SOM), which can derive the multi-templates for each category from input data. The fine-classifiers are realized using multilayered neural networks, each of which solves the two-category classification problem. We also use the rough-classifier for the selection the training samples in the learning process of multilayered neural networks 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.