Face Recognition Based On Bayesian Statistical Methods

Face recognition technology is developed as an active research in the field of applied mathematics and pattern recognition. It's broadly applied, such as in security check, law surveillance, and so on. A fully automatic face recognition system contains face detection, feature extraction and recognition. This paper discusses a partly automatic face recognition system which not contains face detection.Principle component analysis is commonly used in face recognition. First, we give a fully theory deduction of principle component analysis, based on the KL transformation. Then we provide a comparison in the four methods of distance-measurement; also recommend an appropriate method in this paper. Second, because of the high-dimensional-complexity calculation of the principle component analysis theory, a new technique method called two-dimensional principal component analysis is demonstrated to be more superior. In this paper, we prove that two-dimensional principal component analysis is not only obviously time-efficient but also highly recognition rate.Bayesian rule is a best method in pattern recognition. We mainly focus on the Bayesian algorithm and the distribution of the model, based on the principal component analysis theory. In the choice of algorithms, maximum likelihood rule is compared with the maximum a posteriori rule. Then we improve the maximum likelihood rule in the complex calculation. It is proved that the improved maximum likelihood algorithm is superior on operational rate than typical principle component analysis method, as well as the recognition rate.We generally choose the typical Gaussian model, which includes Single-Gaussian distribution model and Mixed-Gaussian distribution model, for the probability similarity estimation in face recognition. Consequently, we give a fully explanation about the basic tenets of the Gaussian model. Meanwhile the probabilistic-reasoning-model is introduced and proved to be good in face recognition.