Face Recognition Based On Manifold Learning

Biometric identification techniques have become an important way of pattern recognition. It in terms of information technology is used to measure and analyze the human profiles. Due to its special merits of flexibility, economy and accuracy, face recognition has a broad application future in biometrics security field. Face recognition can be divided into three steps as follows: face detection, feature extraction and classification. As the high dimension of the face image, the Image Processing is very complex. And feature extraction becomes one of the key steps in face recognition. In this paper, we mainly discuss the feature extraction and propose two new methods.The main work of this paper as follows:1. Owing of category labels contain very important information, we put category labels of the training samples in the constraints, and propose a novel learning approach , called Local Preserve Discriminant Projection (LPDP) based on Locality Preserving Projection. This method solves the problem that LPP is an unsupervised algorithm. The experimental results show that the new method is so much the better.2. This paper, a more efficient, accurate, and stable method is proposed to solve the "small sample size problem" based on Two-Dimensional PCA and Neighborhood Preserving Embedding. And we verify the effectiveness of Enhance Two-Dimensional Neighborhood Preserving Embedding (E2DNPE) by experiments on ORL and Yale database.3. Using PCA for reference, and making use of the percentage of loss function to calculate the percentage loss of mapping matrix in NPE algorithm, Divisor of Numbers (DON) is proposed to resolve the problem of selecting the eigenvectors. In the classification stage of the face feature, we use the nearest neighbor linear classifier and two-dimensional nearest neighbor classifier in LPDP and E2DNPE respectively. Experiments on ORL and Yale face database indicate that the proposed method is more effective than other classical algorithms.