偏旁部首和笔画特征混合的离线中文笔迹鉴别  

Offline Chinese handwriting identification based on partial radicals and stroke feature

作　　者：贾建忠[1] JIA Jian-zhong(Urumqi Vocational University,Urumqi 830001,China)

机构地区：[1]乌鲁木齐职业大学,乌鲁木齐830001

出　　处：《信息技术》2020年第8期60-64,共5页Information Technology

基　　金：教育部青年基金资助项目(15YJC880028)。

摘　　要：为了提高离线中文笔迹鉴别的识别率和稳定性,提出了一种混合式特征提取方法。方向性笔画曲率特征提取方法具有文本无关性强、处理简单等特点,但全局特征提取不足。基于偏旁部首的网格笔画像素密度特征具有全局特征提取充分,鲁棒性强等特点,但对笔画细节特征的提取不够。文中方法提取以上两种特征并设置合适权重组成混合特征向量进行距离测度。实验结果表明,混合两种特征提取方法可以显著提高笔迹鉴别的识别正确率及对不同样本的识别效果稳定性。In order to improve the recognition rate and stability of off-line Chinese handwriting identification,a hybrid feature extraction method is proposed.The curvature feature extraction method of directional stroke has the characteristics of text independence and simple processing,but the stability of recognition is not strong.The pixel density feature of grid strokes based on partial radicals has the characteristics of good recognition stability and strong robustness,but the extraction of stroke details is not enough.This method extracts the above two features and sets the appropriate weights to form the mixed feature vector for distance measurement.The experimental results show that the hybrid two feature extraction methods can significantly improve the stability of handwriting recognition and make the recognition accuracy slightly higher than the single method.

关 键 词：笔迹鉴别 特征提取 笔画曲率 稳定性 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]