SVM多类分类算法及其在手写体数字识别中的应用  

SVM Multi-class Classification Algorithm and its Application in Handwritten Numeral Recognition

作　　者：李云峰[1] 胡文平[1] 

机构地区：[1]中国电子科技集团公司第二研究所,山西太原030024

出　　处：《物流工程与管理》2012年第7期131-134,共4页Logistics Engineering and Management

摘　　要：SVM(支持向量机)的多类分类是近年来模式识别领域的热门方向。文中描述了一种将多类分类问题转化为两类分类问题的方法,把该方法同PCA(主成分分析)和基于核方法的两类分类方法相结合,生成了一种新的SVM多类分类算法。基于该方法,文中设计了完整的手写体数字识别算法,并使用手写体数字数据集对所提出的算法进行了测试。结果表明,识别全过程的时间复杂度有所降低,识别率可达到85.7%。SVM （ support vector machine ） for multi - class classification is in recent years the field of pattern recognition of the popular direction. This paper describes a multi class classification problem into a category two classification methods, the method with the PCA （ principal component analysis ） and two class classification based on kernel method is combined with the method of, generating a new SVM classification algorithm. Based on this method, I designed a complete handwritten numeral recognition algorithm, and the use of handwritten digital data sets of the presented algorithms were tested. The results show that, the time complexity of the recognition process is decreased,the recognition rate can reach 85.7%.

关 键 词：两类分类 多类分类 核函数 PCA(主成分分析) 支持向量机 

分 类 号：O142[理学—数学]