Handwritten character recognition using learning pattern generation by nonlinear normalization

摘要

In pattern recognition system, the learning pattern size is restricted in many cases. If the learning pattern size is small, an error is occurred in the presumed pattern distribution. As its result, the recognition accuracy is decreased. In this paper, we propose the method for generating variety of learning patterns which have the feature of handwritten characters by using nonlinear normalization technique conversely. We show an increase in recognition accuracy by experiment using the database ETL9B (71 letters of hiragana). In addition, we are succeeded to increase recognition accuracy by selecting generated adequate patterns.