Facial feature extraction based on principal component analysis and class independent kernel sparse representation

Abstract

Robust Principal Component Analysis (RPCA) and kernel sparse representation technology which have been proposed in recent years provide a new idea for solving problems of the above three aspects. In this Thesis, classification algorithm of kernel sparse representation has been proposed based on robust principal component analysis by using RPCA technology to generate redundant dictionary and kernel sparse representation to structure classifier, and has been used for face recognition. Basic idea of this algorithm is to generate base dictionary and error dictionary by using RPCA technology, and to realize face recognition through classifier structured by kernel sparse representation. Firstly, each training sample matrix has been decomposed into a low rank matrix and a sparse error matrix by using RPCA technology, so as to structure base dictionary and error dictionary by using the low rank matrix and error matrix respectively, and generate redundancy dictionary of sparse representation of test samples. Then, Kernel regularized Orthogonal Matching Pursuit (KROMP) algorithm has been proposed to get sparse representation coefficient which has been used to complete classification and recognition of test samples. Compared with similar algorithms, algorithm in the Thesis is of a high recognition rate for face recognition, and has a strong ability to adapt to noise and error interference.