摘要
针对人脸识别中的特征提取问题,对原始的非监督判别映射(UDP)算法进行了改进,提出一种基于核正交UDP的人脸识别算法.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;在加入基向量正交的约束后,通过能够保持人脸图像局部几何结构的UDP算法作一个线性映射,以求取算法的正交基向量.该算法中,采用核方法可以更好地提取人脸非线性结构特征,正交基向量则可以更好地保留非线性子流形空间与度量结构有关的信息,增强了算法的识别性能.最后,通过在ORL和PIE人脸库上的人脸识别实验验证了文中算法的有效性.
In view of the problems of feature extraction in face recognition,an improved version of unsupervised discriminant projection（UDP） named kernel orthogonal unsupervised discriminant projection is proposed in this paper.First the nonlinear information in face images is extracted by the kernel trick and mapped into a high dimensional nonlinear space.Then a linear transformation which produces orthogonal basis vectors is performed to preserve locality of the geometric structure of the face images.The kernel trick helps obtain nonlinear structure features and the orthogonal basis vectors help preserve the information of nonlinear sub-manifold space related to the metric structure.Experiments on ORL and PIE face database demonstrate the effectiveness of the proposed algorithm.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2010年第10期1783-1787,共5页
Journal of Computer-Aided Design & Computer Graphics
基金
国家"八六三"高技术研究发展计划(2009AA04Z215)
国家自然科学基金(60803036)
关键词
子空间
非监督判别映射
流形
核正交非监督判别映射
人脸识别
subspace
unsupervised discriminant projection
manifold
kernel orthogonal unsupervised discriminant projection
face recognition
