95 articles 

inria-00103732, version 1

Online Recognition of Unconstrained Handwritten Japanese Text Using Statistical Information

Katsumi Marukawa () 1, Takeshi Nagasaki () 1, Kazuyoshi Kikuta () 2

Tenth International Workshop on Frontiers in Handwriting Recognition (2006)

Abstract: We developed an online recognition system to read unconstrained handwritten Japanese texts using statistical information. The substantial problems of reading handwritten Japanese text are how to correctly segment freely written characters and how to correct errors of character classification. Our method searches for the best interpretation by integrating the likelihoods of character segmentation, character classification and language processing. For the language processing, candidates of words in any position are extracted from a dictionary including about 240,000 words, and the extracted words are evaluated using their grammatical connective probability and word bi-gram probability as the context. Experiments using 467 texts showed that our complete method is more accurate than any partial method. The rate of recognition per text improved from 10 to 38%, and the rate of recognition per character improved from 72.2 to 82.1%. The effectiveness of our method was proved.

  • 1:  Central Research Laboratory (CRL)
  • Hitachi Ltd
  • 2:  Government & Public Corporation Information Systems Division (GPCISD)
  • Hitachi Ltd
  • Domain : Computer Science/Document and Text Processing
    Computer Science/Computer Vision and Pattern Recognition
  • Keywords : Online recognition – unconstrained handwritten Japanese text – language processing – grammatical connective probability – word bi-gram probability
  • Comment : http://www.suvisoft.com
  • inria-00103732, version 1
  • oai:hal.inria.fr:inria-00103732
  • From: 
  • Submitted on: Thursday, 5 October 2006 11:10:22
  • Updated on: Thursday, 5 October 2006 11:19:53