Face Recognition Based On The Linear Subspace Method

The automatic recognition of human faces is an active subject in the pattern recognition area over the past few years, which has a wide range of potential applications. Background of the automatic face detection and recognition system project, this paper aims to finish the face recognition task based on the face detection.Firstly, the background, significance and content of this research are summarized. Then, typical methods of face recognition are studied, especially the linear subspace methods and feature classifiers. The paper analyzes the basic theory of Locality Preserving Projection (LPP) and its relationship with Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) in detail. The paper also points out the non-linear feature selection problem of the LPP subspace which could lead to the failure of using nearest neighbor classifier. Therefore, a face recognition method with both LPP subspace and Radial Basis Function (RBF) classifier is proposed according to the advantage of RBF that could convert non-linear separable problems to a linear separability. In short, the process of our method has three steps. First, calculate the face subspace from the training samples by using LPP; then, the new face image to be identified is projected into the subspace; finally, the new face image is identified by the RBF classifier. The paper uses LPP in supervised mode and trains the hidden center of RBF network with Orthogoral Least Squares (OLS) method.A lot of experiments are performed on ORL, Yale-A, Yale-B&Yale-B Extended, Harvard, CMU-PIE and CAS-PEAL face databases. The results show that the recognition rate of our improved method is better than Eigenfaces, Fisherfaces and Laplacianfaces.