¡¡Chinese Journal of Computers   Full Text
  TitleA Novel Hidden Markov Model and Its Application to Recognize Hand-Drawn Graphic Symbol
  AuthorsPEI Ji-Hong1) LI Cui-Yun2) GONG Xin3)
  Address1)(Modern Educational Technology & Information Center, Shenzhen University, Shenzhen 518060)
2)(School of Electronic Engineering, Xidian University, Xi¡¯an 710071)
3)(Nortel Networks (China) Limited, Beijing 100006)
  Year2005
  IssueNo.10(1745¡ª1752)
  Abstract &
  Background 		A novel Hidden Markov Model (HMM) structure for recognizing feature sequence, an adaptive HMM (AHMM) structure, is presented. Distinguished from traditional forward HMM, The AHMM is a closed loop feedback HMM recognition structure. Firstly, a feature compression method, which has a changeable factor used to adjust the feature compression ratio, is applied to compress feature vectors with a high compression ratio. Next, compressed feature sequence is supplied to HMM recognition system. If recognition result is not satisfaction, the compression ratio factor is changed and compress feature vectors are constructed again to obtain longer feature sequence. New feature sequence is supplied to HMM recognition system until a satisfactory recognition is obtained. Superior performance over traditional HMM recognition system is testified by sets of experiment of free form hand-drawn graphic symbol recognition based on AHMM, in which single-band integral algorithm is used to compress feature vectors adaptively according to geometry feature of input graphics symbol.

keywords HMM; feedback; adaptive; feature compression; hand-drawn graphic symbol

background This work is sponsored by National Nature Science Foundation of China. The project name is ¡°Study of a Constrained Adaptive Technique for Online Hand-Drawn Graphics Recognition¡±. The project studies the adaptive recognition technique on online hand-drawn graphic symbol under the constraint that graphic symbols are constructed by basic graphics tokens. Before this project, the research term has accomplished relative project of national defense pre-research project named by ¡°Study to Human-Computer Interface Information Technique¡±. This paper¡¯s work is emphasis on how to recognize hand-drawn basic graphics token adaptively.