Research Of Supervised Manifold Learning In Face Recognition Applications

Today, with the rapid development of information based society. Information security becomes extraordinarily important. The biological feature identification technology is the computer security technology which rapidly expanding all over the world recently, it makes full use of inherently biological feature of human body to identify people, with features of high reliability and stability, and has widely applied areas.As a kind of very important biological feature identification technology, face recognition has high values of study and wide prospect of application from such feature of user friendly, easy of access, not easy to forge and so on. However, due to the reliability of human facial features are influenced by small changes of expression, action, illumination, background and so on, the researches of face recognition technology mainly stay on developing the algorithm of feature extracting, with the hope of extracting sound features and meanwhile reducing the dimension of face images, which will make recognition results more accurate.Base on the thorough research and comparison of the existing face recognition technology at home and broad, we have proposed new face recognition methods based on supervised manifold learning. To be concrete, the main contributions of this paper conclude: first, we give a novel global based face recognition algorithm. Our method combines manifold learning based algorithms with known class information, for preserving the neighborhood structure of facial manifold and meanwhile maximizing the margin between classes in the process of finding feature space through discriminant analysis based method, which is effective for classification. We also give an effective solution to the eigenvalue problem. Our method can avoid the preprocessing stage of resizing the original image resolution and Principle Component Analysis (PCA) projection, so there is no information lost. Second, we give a novel local matching based face recognition algorithm. To the best of our knowledge, this is the first method which incorporates the supervised manifold learning based method into local matching face recognition technique and produce better recognition results compare to other methods. In a word, the proposed face recognition methods in this paper demonstrate the feasibility and efficiency by experiments.