基于SLPP和张量分解相结合的人脸识别*

针对多线性分析算法对多姿态多身份因素并存时，人脸的识别率大大下降等问题，提出了带监督的局部保留投影映射算法与多线性张量分析算法相结合的人脸识别方法。该方法将人脸转动的近邻点信息作为监督信息引入，更精确地描述了姿态空间的非线性结构，再结合张量分解和核函数将姿态流形系数映射到高维图像空间，使得从低维空间到高维空间映射的精确性得以提高。在东方人脸数据库上进行实验，结果验证了该算法的有效性。

Supervised locality preserving projection and tensor decomposition for multi view face recognition

To solve the incapability of multi-linear analysis algorithm with multi-view face images, this paper proposed an improved algorithm . A supervised locality preserving projection that accurately described the nonlinearity in pose embedding space was introduced to project the face images into a low dimensional subspace. In the SLPP algorithm the neighboring points as supervised information could find the global structure as well as local structure. Combining with tensor decomposition and kernel methods, it well learned a set of nonlinear mapping functions from the embedding space into the input space. The proposed method was evaluated on Oriental Face database of multi-view face images in comparison to the other subspace methods. Experimental results show the effectiveness of the new method.