構造化字体表現型HMMに基づくオンライン手書き文字認識における座標特徴の利用法と効果  [in Japanese] Pen-Coordinate Information Modeling by SCPR-based HMM for On-line Japanese Handwriting Recognition  [in Japanese]

Search this Article

Author(s)

Abstract

オンライン手書き日本語文字認識を対象に,漢字の扁や旁などのサブパターンを共有する構造化字体表現型HMMにおいて,文字パターンの方向特徴に加えて座標特徴をモデル化する手法とその効果を述べる.構造化字体表現型HMMでは,HMMの観測信号として方向特徴を用いるのが一般的であり,文字依存性の強い座標特徴をモデル化する手法は報告されていない.これに対し,本研究では学習データから抽出したサブパターンに対する線形写像および,文字の構造情報に基づいたHMMのパラメータ適応を行うことによって,構造化字体表現HMMで座標特徴をモデル化できること,および,その効果を示す.

This paper describes stochastic modeling of pen-coordinate information in HMMs with structured character pattern representation (SCPR) for on-line Japanese handwriting recognition. SCPR allows HMMs for Kanji character patterns to share common subpatterns. Although SCPR-based HMMs have been successfully applied to Kanji character recognition, the pen-coordinate feature has not been modeled since it is unique feature in each character pattern. In this paper, we employ mapping from a common subpattern to each occurrence in Kanji patterns in the estimation step of SCPR-based HMMs. We also employ adaptation of state parameters to each character pattern in generating character HMMs by composing SCPR-based HMMs. Experimental results show that the pen-coordinate feature modeled in the SCPR-based HMMs effects significantly.

Journal

  • IEICE technical report

    IEICE technical report 105(613), 117-122, 2006-02-24

    The Institute of Electronics, Information and Communication Engineers

References:  12

Codes

  • NII Article ID (NAID)
    110004662938
  • NII NACSIS-CAT ID (NCID)
    AN10449078
  • Text Lang
    JPN
  • Article Type
    ART
  • ISSN
    09135685
  • NDL Article ID
    7848865
  • NDL Source Classification
    ZN33(科学技術--電気工学・電気機械工業--電子工学・電気通信)
  • NDL Call No.
    Z16-940
  • Data Source
    CJP  NDL  NII-ELS 
Page Top