基于2DPCA的手写数字识别  

Handwritten digit recognition based on 2DPCA

作　　者：王军平[1] 赵振华[2] 

机构地区：[1]咸阳职业技术学院电子信息系,陕西咸阳712000 [2]兰州理工大学电气工程与信息工程学院,甘肃兰州730050

出　　处：《电子设计工程》2012年第21期58-61,共4页Electronic Design Engineering

摘　　要：手写字符由于书写风格和习惯的不同,造成字符模式的不稳定。针对这一问题,本文首先对字符图像进行图像预处理,统一字符笔画的粗细,改善局部特征,随后利用二维主分量分析法(2DPCA)直接对字符图像矩阵进行变换,抽取字符特征,建立字符的特征矩阵及重构模型,利用最邻近方法和重构误差法进行字符识别。最后通过美国国家邮政局MINIST手写数字库中进行识别实验,验证了算法的准确性和鲁棒性。To overcome the problem of handwritten numerals models instability caused by different writing styles, a robust method is presented in this paper. First, the numeral image was pre-processed to be of similar thickness and improved for its local features. Then 2-dimension principal component analysis （2DPCA） is directly used for character image to extract character features and establish character＇s feature matrixes and reconstruction models. Character recognition is conducted based on a nearest neighbor classifier and reconstructed models＇ error method respectively. Finally, the algorithm proposed in this paper is tested on all the characters in MINIST database and a comparison is made with one dimension principal component analysis, and the experimental results validate the accuracy and robustness accuracy of the proposed algorithm.

关 键 词：特征矩阵 重构 最邻近法 2DPCA 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]