Face Recognition Based On Subspace

Personal identification is necessary to ensure the security of the system. The biologic character is a kind of widespread and relatively difficult to forge, Face recognition is the most natural and friendly biometric identification technology and has strong academic research value and application value.Subspace face recognition is the main method for face recognition. The main idea is to project face images of high dimensional to the fit subspace, the classification and recognition, classify and recognize. Subspace method basically divided into two categories, the expression subspace method and the identification subspace method. Face is a potential low-dimensional manifold in high-dimensional space. Compared to the dimension of face image samples, the number and face image is relatively small. The face recognition problem is the high dimensional data problem and the small training sample size problem.(1)Aiming at the high dimensional data problem, in this paper, we combined with principal component analysis and independent component analysis, to use the order statistics. To improvement efficiency, realize two-dimensional principal component analysis and independent component analysis. Experiment results show that the two methods are better than the principal component analysis or independent component analysis alone.(2) Aiming at the small sample problem, in this paper, we propose the method self-training for margin neighbor. Use the margin to represent decision confidence, use the spatial adjacent to represent gradual change of face manifold, through self-training iteration, the sample distance of the same classifications is as compact as possible, the sample distance of the different classifications maintain a certain large distance. In the neighborhood, constantly mark the unlabeled samples of high credibility.The algorithm was improved twice. First, according to the results, add the data editing to the self-training, to remove as much as possible the bad impact of boundary samples or outliers. According to the algorithm complexity, enhance neighbor constraints, in addition to a neighbor computing. Experiments show that, compared to other methods, continuous improved self-training for large margin neighbor has relatively better recognition in small face samples.