|
Pattern recognition is the rapid development of a new subject in the past 30 years, the main purpose is to study how to use the machine to simulate human learning, recognition and thinking ability. Visual image based pattern recognition technology is widely used in industrial, commercial, agricultural, military, medical and other fields. Pattern recognition of a biometric authentication technology (Biometrics) spent inherent human biometric new technology is completely different with the traditional authentication methods, having a higher level of security, reliability, and effectiveness, the more and more attention has been paid. In a wide range of authentication methods, face recognition technology is one of the most easy to accept identification method, in recent years, computer vision, pattern recognition is a major research focus, solve the case in the criminal investigation, document authentication, access control systems, video monitoring and other fields have a wide application prospect. This paper first introduces the research background and the main method of face recognition technology, face recognition technology - a key link in facial feature extraction and classification depth study, proposed a set of static face recognition method . In-depth analysis of the PCA (Principal Component Analysis) and KPCA (Kernel Principal Component Analysis) method on the basis of this paper, a two-stage nuclear feature extraction method: PCA KPCA, during the non-linear mapping, the first to use the classic Principal Component Analysis dimensionality reduction, and then run the kernel principal component analysis (KPCA). The test results on the ORL face database to verify the effectiveness of the proposed algorithm. The same time, the use of two-stage feature extraction on the basis of law, building a framework for face recognition. Extracted features two-stage method with BP artificial neural network training and recognition. Neural network training, for the BP neural network is sensitive to the initial weights, easy to fall into local minima, this paper presents a global search ability of swarm intelligence optimization method - particle swarm optimization algorithm (PSO) is given close to the global minimal BP neural network initial weights, on this basis, then using the momentum method and adaptive learning rate adjustment strategy to improve BP algorithm. The framework of the two-stage method optimization feature extraction and PSO-BP neural network combined, achieved a higher recognition rate and speed, there is a certain degree of anti-noise performance. Experiments show that the frame of discernment on the ORL face database is accurate and effective, successful experiment.
|