Improvement of Learning algorithm for Locally Connected Neural Networks and Three-Stage Recognition Method of Hand-Written Characters.

  • Ohtomo Teruhiko
    Department of Basic Technology, Faculty of Engineering, Yamagata University
  • Yang Qing
    Department of Basic Technology, Faculty of Engineering, Yamagata University
  • Otsuki Takashi
    Department of Basic Technology, Faculty of Engineering, Yamagata University

Bibliographic Information

Other Title
  • 神経回路網のモデルにおける学習アルゴリズムの改良と手書き文字の3段階認識法

Search this article

Description

For hand-written character recognition system based on neural network model, we have the following problems : a) the most suitable structure of network, b) the diminution of learning time, c) effective treatment for similar characters, d) generality. From those viewpoint, we proposed a 3-stage hand-written character recognition system which use locally connected neural network models. Here, we propose a new δ-learning method which is based on back-propagation algorithm. By using those techniques, we obtained 96.1% recognition rate for 200 Kanji types (8000 patterns which were not used for learning), and 92.0% recognition rate for 4000 patterns of diferent writers.

Journal

Keywords

Details 詳細情報について

  • CRID
    1573950402107830784
  • NII Article ID
    110003299479
  • NII Book ID
    AN10013232
  • Text Lang
    ja
  • Data Source
    • CiNii Articles

Report a problem

Back to top