中文摘要：

针对人脸识别中的非线性特征提取问题,基于邻域保持嵌入,提出了一组在核空间具有正交性鉴别矢量和一组在核空间具有统计不相关性鉴别矢量的计算方法.算法首先利用核的方法提取人脸图像中的非线性信息,并将其投影在一个高维非线性空间.然后在目标函数中最小化核空间类内邻域散度并最大化核空间类间邻域散度来增强算法的分类鉴别能力.最后通过引入基向量正交和不相关约束得到核正交及核不相关邻域保持鉴别嵌入算法,并给出求解2种算法基向量的一般性定理及其推导过程.Yale和PIE库上的人脸识别实验验证了算法的有效性,实验结果表明算法能有效降维并提高鉴别能力.

英文摘要：

In view of the problems of nonlinear feature extraction in face recognition,a new algorithm of orthogonal optimal discriminant vectors and a new algorithm of statistically uncorrelated optimal discriminant vectors in a kernel space were proposed based on neighborhood preservation embedding（NPE）.First,nonlinear kernel mapping was used to map the face data into an implicit feature space.Then the algorithm maximized inter-class neighborhood scatter information while minimizing intra-class neighborhood scatter information in the kernel space,which helped to improve its discriminant ability.Finally,the kernel orthogonal preserving discriminant cmbedding（KONPDE） algorithm and the kernel uncorrelated neighborhood preserving,discriminant embedding（KUNPDE） algorithm were obtained by constrainting the base vectors orthogonal and uncorrelated respectively. Also the general theorem for solving the base vectors of the above two algorithms and the derivations of the algorithms were specifically introduced.Experiments on Yale and PIE demonstrate the effectiveness of the algorithms,and show that these algorithms can reduce the dimensions of the data and improve the discriminant ability.