模式分类器在手写体数字识别中的应用比较研究  

Comparison of Pattern Classifiers on Handwritten Digit Recognition

作　　者：汪青[1] 干静[1] 

机构地区：[1]四川大学制造科学与工程学院,四川成都610065

出　　处：《装备制造技术》2008年第8期111-114,共4页Equipment Manufacturing Technology

摘　　要：贝叶斯分类器、线性分类器和K近邻分类器是模式识别中三种典型的模式分类器。比较三种分类器在识别手写阿拉伯数字过程中的性能优缺点,进一步对识别数据进行详尽的分析挖掘,通过对算法精确度、识别速度及计算存储需求等方面的比较,深入探讨三种监督式分类器的差异和特点,最终得到不同的分类结果,从而寻求最优化的决策方案。Bayes classifier, linear classifier and K-nearest neighbor classifier are the typical supervised learning classifiers in pattern classification. In this paper, the advantages and disadvantages of these classification rules are discussed in terms of handwritten digit recognition problem. We explore the handwritten digit database and compare the performance of different classifiers in terms of error rate and computation cost. The results show that the error rate is lower for the bays rule and the K-NN than that of Linear classifier; however, the linear classifier is easier to implement and widely used.

关 键 词：手写数字识别 贝叶斯决策 K近邻决策 线性分类器 误差率 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]