基于字典学习与稀疏表达分类的低质量字符识别  

Low-Quality Characters Recognition Based on Dictionary Learning and Sparse Representation

作　　者：郝宁波[1,2] 廖海斌[3] 杨杰[1] 

机构地区：[1]武汉理工大学信息工程学院,湖北武汉430070 [2]黄淮学院国际学院,河南驻马店463000 [3]湖北科技学院计算机科学与技术学院,湖北咸宁437100

出　　处：《华南理工大学学报（自然科学版）》2016年第5期123-129,共7页Journal of South China University of Technology(Natural Science Edition)

基　　金：中国博士后科学基金面上项目(2015M582355);公安部科技攻关项目(SN20110001)~~

摘　　要：为解决低质量字符中的断笔、噪声和模糊问题,以及不同字体与字号的字符识别问题,提出了基于字典学习与稀疏表达分类的低质量字符识别方法.首先,收集不同字体和字号的字符样本构建字符超完备字典;然后,对测试字符进行稀疏表达建模,并根据求解的稀疏系数进行字符分类.为了使字典更具鉴别性,文中提出了基于因子分析的字典学习方法.实验结果表明,文中所提方法不仅可以同时识别不同字体和字号的字符,还具有对断笔、噪声和模糊的鲁棒性.In order to recognize low-quality characters with interrupted strokes,noise and fuzziness,and to recognize characters with different fonts and sizes,a method to recognize low-quality characters on the basis of dictionary learning and sparse representation is proposed. Firstly,character samples with different fonts and sizes are collected to construct a super-complete dictionary of characters. Then,a sparse representation model is established by using test characters,and a character classification is made according to the solved sparse representation coefficient. Additionally,in order to make the dictionary more discriminating,a dictionary learning method on the basis of factor analysis is proposed. Experimental results show that the proposed method not only can identify characters with different fonts and sizes but also possesses robustness to interrupted strokes,noise and fuzziness.

关 键 词：字符识别 字典学习 稀疏表达 因子分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]