Research On Face Recognition System Based On BP Neural Network

Face recognition has become a research focus in the field of pattern recognition and digital image processing recently. As one of several important human external characteristics, face recognition has both significant theoretic values and wide potential applications for automatic figure verification and identification. In this paper, the main methods for face detection and segmentation, as well as the improvement and selection of algorithms of BP neural network are studied and summarized. This paper focus on the improvement of algorithms for face feature extraction, and the design and realization of BP neural network based face image recognition system. The main studies are as follows:Firstly, in the method of the research of the face feature extraction and selection, based on Z_l-Z_r, it computes J criterion of features, and then adds or subtracts some features for getting a better answer. Meanwhile, to resolve the problems of the feature correlation and the complication of the computing, K-L transform is used for the vector combined from n-dimension original features in order that the features have no correlation in new space. When the method reduces the complication of the feature selection's computing, it can adjust the parameters to improve its practicability.Secondly, a face image recognition system blue print based on BP neural network is designed. The system structure and system flow is designed, and divided the system into four modules including image collection, face detection and segmentation, feature extraction and selection and face recognition. And, each module's function and the data processing mode is defined. Then, the design and realization of the face recognition module is presented mainly. In this module, a improved BP neural network is used as core method. Compared with other method, this method has certain improvement in recognition rate and stability.