基于投票极限学习机的人脸识别混合算法研究  

Research on Hybrid Face Recognition Algorithm Based on Voting Extreme Learning Machine

作　　者：陈铁军[1] 蔡金收 郭丽[1] 

机构地区：[1]郑州大学电气工程学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2016年第2期37-41,共5页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(41174127);教育部博士点基金资助项目(20114101110005)

摘　　要：针对小波分析在处理多维图形时不能充分利用数据本身特有几何特征的缺陷,使用第二代曲波变换(the second generation of curvelet transform,SGCT)方法进行人脸图像的处理,选取具有最大标准差的尺度层系数,以完成对人脸图像的特征提取,同时结合基于双向二维主成分分析(bidirectional two dimensional principal component analysis,B2DPCA)的数据降维,构造一种基于混合投票机制极限学习机(voting extreme learning machine,VELM)的人脸识别算法.通过与其他算法的分类结果对比,证明该算法具有更高的识别正确率.In view of the deficiency of wavelet transform that geometrical characteristic of the data can not be fully atilized when multidimensional images＇ features are detected,this paper applies the second generation of curvelet transform to process facial images. The procedure extracts images＇ features through choosing the subband coefficient with the maximum standard deviation. Then data dimensionality is reduced using the bidirectional two-dimensional principal component analysis（ B2DPCA）. Combining with the two methods above,this paper constructs a face recognition algorithm based on hybrid voting extreme learning machine（ VELM）. The experimental results show that the algorithm presents improved performance compared with other algorithms mentioned in terms of recognition accuracy.

关 键 词：曲波变换 B2DPCA 投票 ELM 人脸识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]