一种有监督的LPP算法及其在人脸识别中的应用

A Supervised LPP Algorithm and Its Application to Face Recognition

为了提高局部保持投影算法(Locality Preserving Projections,LPP)对光照、姿态等外部因素的鲁棒性,该文对传统的LPP算法进行改进,提出了一种有监督的LPP(SLPP)方法。首先对LPP子空间进行判别分析,然后选择主要反应类内差异的基向量来构造子空间,最后在子空间上进行识别。通过Havard人脸库和Umist人脸库上的实验,结果表明该方法能够对光照和姿态的变化具有一定的鲁棒性和较高的识别率,比传统的LPP方法和其它子空间分析法识别率提高了10%以上。

Illumination and pose variations make the performance of the Locality Preserving Projections （LPP）in face recognition decrease. To solve the problem, a supervised LPP using discriminant information is presented in this paper, the proposal calls for the establishment of a feature subspace in which the intrasubject variation is minimized, while the intersubject variation is maximized, then face recognition is implemented with the subspace. Experimentation results on Havard and Umist indicate that this approach is robust to illumination and pose and has higher recognition rate than LPP and other subspace methods.