Face Recognition Algorithm Based On Statistical Research

Twenty years ago, the problem of face recognition was considered among the hardest in artificial intelligence and computer vision. Surprisingly, however, over the past decade, with the development of science technology and the demand of commercial, law, smart environment and wearable computers enforcement applications, face recognition technology has received significant attention again.In this paper the development of face recognition technology is firstly introduced, then the problems of feature extraction, feature selection and classification are discussed. After analyzing the application of Bayesian Optimization Algorithm (BOA), a new randomized, population-based evolutionary algorithm is proposed which deals with the Eigenvector Subset Selection (ESS) problem on face recognition application. The proposed algorithm is called the ESS-BOA. Experimental results show that ESS-BOA can represent features of faces optimizlly. It outperforms the traditional eigenface selection algorithm.Secondly the classification methods for face recognition are discussed, a new classification method-the nearest feature line(NFL) method is deeply studied. After analyzing the validity of NFL for face recognition in theory and experiments, the K-NFL method is introduced in face recognition. Experimental results show that K-NFL fits face recognition too. However these two methods have a common disadvantage, that is higher computation complexity. So here the valid solutions of the problem are discussed too.Finally the four traditional distance measures in the context of face recognition are discussed. And a new distance measure called weighted Mahalanobis is proposed. Experimental results of PCA-based face recognition show that the weighted Mahalanobis distance measure fits face recognition and improve the recognition rate.