Segmentation-driven offline handwritten Chinese and Arabic script recognition

  • Authors:

  • Xiaoqing Ding;Hailong Liu

  • Affiliations:

  • Dept. of Electronic Engineering, Tsinghua University, State Key Laboratory of Intelligent Technology and Systems, Beijing, China;Dept. of Electronic Engineering, Tsinghua University, State Key Laboratory of Intelligent Technology and Systems, Beijing, China

  • Venue:
  • SACH'06 Proceedings of the 2006 conference on Arabic and Chinese handwriting recognition
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

The market of handwriting recognition applications is increasing rapidly due to continuous advancement in OCR technology. This paper summarizes our recent efforts on offline handwritten Chinese script recognition using a segmentation-driven approach. We address two essential problems, namely isolated character recognition and establishment of the probabilistic segmentation model. To improve the isolated character recognition accuracy, we propose a heteroscedastic linear discriminant analysis algorithm to extract more discrimination information from original character features, and implement a minimum classification error learning scheme to optimize classifier parameters. In the segmentation stage, information from three different sources, namely geometric layout, character recognition confidence, and semantic model are integrated into a probabilistic framework to give the best script interpretation. Experimental results on postal address and bank check recognition have demonstrated the effectiveness of our proposed algorithms: A more than 80% correct recognition rate is achieved on 1,000 handwritten Chinese address items, and the recognition reliability of bank checks is largely improved after combining courtesy amount recognition result with legal amount recognition result. Some preliminary research work on Arabic script recognition is also shown.