基于SVM的几种汉字特征提取法比较研究  

Study of Feature Extraction for Several Chinese Characters Based on SVM

作　　者：肖斌[1] 黄襄念[1] 

机构地区：[1]西华大学数学与计算机学院,四川成都610039

出　　处：《西华大学学报（自然科学版）》2009年第5期70-74,共5页Journal of Xihua University:Natural Science Edition

摘　　要：阐述了边缘法、骨架法以及笔画法的方向特征提取算法的基本思想以及支持向量机(SVM)基本原理,采用SVM对所述特征提取技术分别得到的特征样本进行分别识别,正确识别率都达到90%以上。然后在时间效率、识别率以及汉字书写风格对特征提取算法的影响三方面的的对比分析基础上,得出笔画特征提取法是本文所述几种特征提取方法中时间效率最高(平均识别时间1.54ms),正确识别率最高(达96.6%)的特征提取方法。This paper introduces several directional features extraction approaches, such as meshing methods including edge approach, framework approach and stroke approach for handwritten Chinese character recognition, and analyzes the principle of SVM （Support Vector Machine）. SVM method is applied to respective feature extracted by every feature extraction approach mentioned above and the correct rate of the recognition is satisfactory （more than 90% ）. At last, on the basis of comparison and analysis of several feature extraction approaches, the conclusion is drawn that the best approach is Stroke-based directional feature extraction in terms of its correct recognition rate （96.6%） and average recognition time （ 1.54ms per character） among these feature extraction approaches.

关 键 词：手写体汉字识别 网格方向特征提取 支持向量机 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]