基于多元正态分布的手写数字识别研究  

Handwritten digits recognition based on multivariate normal distribution

作　　者：叶晓波[1] 秦海菲[1] 吕永林[1] Ye Xiaobo;Qin Haifei;Lu Yonglin(Chuxiong Normal University,Chuxiong,Yunnan 675000,China)

机构地区：[1]楚雄师范学院,云南楚雄675000

出　　处：《计算机时代》2018年第10期11-14,18,共5页Computer Era

基　　金：云南省教育厅科学研究基金项目(2012Y131)

摘　　要：介绍了多元正态分布及边界判别方法,详细阐述了数据降维方法中的主成分分析(PCA降维),分析了MNIST手写数字数据库的数据结构和数据读取方法,论述了基于多元正态分布的手写数字识别方法的整个实验过程。实验结果显示,经PCA降维后的手写数字呈现多元正态分布的特征,使用正态分布的边界判别方法对手写数字进行识别运行效率高。A handwritten digits recognition method based on multivariate normal distribution is proposed and studied experimentally. This paper introduces the multivariate normal distribution and boundary discrimination method, expounds in detail the principal component analysis （PCA） in the data dimension reduction method, and analyzes the data structure and data reading method of the most popular handwritten digits database MNIST. The whole experiment process of handwritten digits recognition method based on multivariate normal distribution is discussed in detail. The experimental results show that the handwritten digits reduced by PCA show- the characteristics of multivariate normal publication. The boundary discriminant method based on normal distribution is efficient to recognize handwritten digits.

关 键 词：手写数字识别 主成分分析(PCA) MNIST 多元正态分布 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]