交通运输系统工程与信息

交通运输系统工程与信息 ›› 2006, Vol. 6 ›› Issue (1): 51-54 .

• 智能交通系统与信息技术 • 上一篇    下一篇

面向连续手写文本的部件识别研究

赵巍，刘家锋，唐降龙，郭延辉   

  1. 哈尔滨工业大学 计算机学院，哈尔滨 150001
  • 收稿日期:2005-12-20 修回日期:1900-01-01 出版日期:2006-02-20 发布日期:2006-02-20

A Continuous-Recognition-Oriented Handwritten Chinese
Radical Recognition Research

ZHAO Wei， LIU Jia-feng，TANG Xiang-long，GUO Yan-hui   

  1. School of Computer, Harbin Institute of Technology，Harbin 100051，China
  • Received:2005-12-20 Revised:1900-01-01 Online:2006-02-20 Published:2006-02-20

PDF

被引次数

7

摘要： 联机连续文本识别是字符识别技术领域中新的研究方向.基于分层构筑法(Level-Building, LB)和动态时间规整算法(Dynamic Time Warping, DTW)建立了面向连续手写文本识别的手写部件识别器。将部件看作笔段和连续文本的中间模式，根据手写文本的特点建立了由484个手写部件构成的部件集.提取笔段的长度、角度等特征用于LB中每一层的DTW网格匹配中.测试样本包括6763个汉字和303个连续手写文本.实验结果表明手写体部件集能够有效地支撑笔段和连续文本之间的联系，串识别率达到86.47%。

关键词: 联机连续手写文本, 手写部件, 对数正态分布, LB与DTW融合算法

Abstract: n this paper, a handwritten radical recognizer, with the purpose of obtaining reliable radicals in online Chinese words or sentences recognition task, was designed based on a hybrid method of Level-Building (LB) and Dynamic Time Warping (DTW) algorithm. Radicals were considered as mid-patterns between strokes and continuous handwritten script in the recognizer. A handwritten radical set was established in terms of handwritten script characteristics. Adjacent stroke relative feature vector sequences were put into the grid point matching process of DTW in each level of LB structure. The test samples include 303 handwritten sequences and 6 763 Chinese characters. It is shown that the radical set could be in effect between strokes and a handwritten sequence, 86.47% of recognition rate is obtained.

Key words: on-line continuous handwritten text, handwritten radicals, Logarithmic normal distribution, hybrid of LB and DTW algorithm