一种基于流形学习重建的脱机手写数字识别算法研究  

Handwritten numeral recognition by Model Reconstruction based on Manifold Learning

作　　者：吴炜[1] 杨晓敏[1] 何小海[1] 陈默[1] 邓波[1] Xiao-Min Xiao-Hai 

机构地区：[1]四川大学电子信息学院图像信息研究所,成都610064

出　　处：《四川大学学报（自然科学版）》2009年第6期1679-1684,共6页Journal of Sichuan University(Natural Science Edition)

基　　金：四川大学青年科学研究资助项目

摘　　要：该文针对在手写字符识别中由于书写习惯和风格的不同造成字符模式不稳定的问题,提出了一种基于流形学习重建的手写体数字识别方法.该方法与传统的流行学习识别方法不同,传统的方法是先对待识别模式进行降维,然后再对降维后的特征进行分类识别;本文提出基于重建的LLE算法(简称RLLE).该算法首先分别对每一类字符训练样本进行LLE降维,得到每一类字符降维后的向量.然后将待识别字符分别在每一类字符中进行降维,并依据该降维后的矢量在每一类中对字符进行重建.最后选取重建误差最小的为待识别的所属的类.通过对MINST字库的一系列实验表明该算法具有较高的识别率,同时该方法为手写体数字识别的研究提供了一条有效的新途径.To overcome handwritten character model's instability caused by different writing style,a novel approach by Model Reconstruction based on manifold learning is proposed in this paper.Traditional method recognition method use LLE to reduce dimension at first,and then recognize the character by reduced feature.The algorithm by Reconstruction first reduce dimension at each class and get the feature vector.Then reconstruct the character according the feature vector.Finally,character recognition is then conducted based on the error analysis of reconstructed models'.The algorithm proposed in this paper is tested on the characters in MINST character database,and the experimental results demonstrate that the method can effectively improve the recognition rate of handwritten digits and provide a new approach to the research for handwritten digits recognition.

关 键 词：流形学习 重建 脱机 手写数字识别 算法研究 Learning based characters 手写字符识别 手写体数字识别 降维 识别方法 MODEL Reconstruction novel approach error analysis writing style paper MODEL LLE different 

分 类 号：N55[自然科学总论]