Dictionary Learning by Adding Hyperspherical Templates

• KIMURA Yoshimasa

  NTT Service Integration Laboratories

• WAKAHARA Toru

  NTT Cyber Solutions Laboratories

• SANO Mutsuo

  NTT Cyber Solutions Laboratories

Bibliographic Information

Other Title

• 有効範囲付テンプレート追加登録法による認識辞書の学習
• ユウコウ ハンイ ツキ テンプレート ツイカ トウロクホウ ニ ヨル ニンシキ ジショ ノ ガクシュウ

Abstract

The traditional adding template method has a serious flaw in that the patterns correctly recognized before becomes to error as dictionary learning progresses. To solve this problem, we present a new adding template method that embeds erroneous pattern as a hypersphere with a small radius into Voronoi region generated per category. Also, we present a hierarchical system that outputs the recognition result from Voronoi regions given by reference templates as the first candidate. Next, hypershperical classifiers embedded in the Voronoi region of the first candidate determine the correct candidate. The hierarchical system promotes the recognition rate by dictionary learning and succeeds in overcoming the serious flaw in experiments on handwritten Kanji pattern recognition.

Journal

• The Journal of the Institute of Image Electronics Engineers of Japan

  The Journal of the Institute of Image Electronics Engineers of Japan 30 (1), 11-20, 2001

  The Institute of Image Electronics Engineers of Japan

Keywords

• Dictionary learning
• Hyperspherical templates
• Hierarchical classification
• Handwritten Kanji character recognition

Details

• CRID: 1390001204612265088
• NII Article ID: 130004437209
• NII Article ID: 10010069870
• NII Book ID: AN00041650
• DOI: 10.11371/iieej.30.11
• ISSN: 13480316
• ISSN: 02859831
• NDL BIB ID: 5734318
• Web Site: https://ndlsearch.ndl.go.jp/books/R000000004-I5734318
• Text Lang: ja
• Data Source:
  • JaLC
  • NDL
  • CiNii Articles
• Abstract License Flag: Disallowed