核的正交完备鉴别局部保持投影  

Orthogonal complete discriminant locality preserving projections of kernel

作　　者：林克正[1] 荣友湖 吴迪[1] 庄靓玮 李鹏[1] 

机构地区：[1]哈尔滨理工大学计算机科学与技术学院,哈尔滨150080

出　　处：《计算机应用研究》2015年第5期1589-1592,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61103149);黑龙江省教育厅科学技术研究项目(11551087)

摘　　要：针对完备鉴别局部保持投影算法所求得的最优判别矢量间存在信息冗余问题,提出了核的正交完备鉴别局部保持投影算法。通过将核函数技术与正交性原理融合,采用高斯核函数将原始样本映射到高维特征空间,在高维特征空间的局部总体散度矩阵中计算最优判别矢量,只需在整个范围内对值域空间进行特征值分解,去除局部零空间达到样本降维目的。该算法分别在UMIST人脸库和JAFFE人脸表情库上进行实验,实验结果表明算法的识别率高达95.59%。This paper proposed the method of the orthogonal complete discriminant locality preserving projections of kernel,which was aimed at the redundant information of the optimal discriminant vectors by the method of complete discriminant locality preserving projections. Combined the convergence kernel technology with orthogonal principle,it mapped the original sample by the method of Gauss kernel function to the high-dimension feature space,and calculated the optimal discriminant vectors in the locality total scatter matrix of high-dimension feature space by the eigenvalue decomposition in the whole range space. It was feasible to remove the locality zero space for the dimensionality reduction. The algorithm made the experiment on UMIST face database and JAFFE facial expression database. Experimental results show that the recognition rate can reach more than95. 59%.

关 键 词：人脸识别 特征提取 完备鉴别局部保持投影 核函数 局部总体散度矩阵 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]