基于ASM算法的特征提取与匹配在文字识别中的应用  

Application of feature extraction and matching based on ASM in OCR

作　　者：孟一飞[1] 张晓彪[1] 杨小花[1] 

机构地区：[1]宁夏大学物理与电子电气学院,宁夏银川750000

出　　处：《广西大学学报（自然科学版）》2017年第6期2183-2190,共8页Journal of Guangxi University（Natural Science Edition）

基　　金：国家自然科学基金资助项目(61162020)

摘　　要：西夏文是西夏王朝创制并使用的文字,笔画繁多,结构复杂,为实现在计算机下的扫描录入,运用改进的ASM算法,对西夏文识别中的笔形特征提取与匹配关键技术进行研究。基于等间距插值方法扩充手工标定的特征点,利用差分和线性插值进行降维处理;采用配准变换和主成分分析(PCA)处理特征点进行形状模型建立;统计特征点局部灰度特征,通过对目标特征点的灰度信息进行收敛,从而实现西夏文笔形的匹配。使用Matlab验证本算法能否有效地将测试笔形匹配到合适的样本集,从而证明该算法的可行性。本算法的研究对于西夏文字的扫描录入有重要的现实意义,并且对各种文字的光学识别具有一定的借鉴价值。The Tangut script,which is originated in the Western Xia dynasty,plays a significant role to the contemporary Tangutology studies. However,its complexity in strokes brings tremendous challenge to the digitalization process. To achieve the character recognition automatically through computer systems,an improved Active Shape Model algorithm is proposed to solve one of the key problems——feature extraction and matching. Firstly,equidistant interpolation is applied to the expansion of feature points,and the expanded points are compressed by using difference equation and linear interpolation. Secondly,registration transformation and Principal Component Analysis are applied to solve the feature points and establish the ASM. Lastly,the gray value of the local features on the target character points are collected, and feature matching is achieved through the convergence of the gray value. The efficacy of the proposed algorithm is demonstrated through tests in Matlab by matching the test strokes images with the right sample set. The proposed algorithm isapplied for optical character recognition of the Tangut script in this study,and it can be broadly used for digital character scanning and recording of other scripts.

关 键 词：西夏文 ASM 特征点 PCA 收敛 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]