Research On Face Recognition Methods Based On Flexible Neural Tree

Face recognition technology is a kind of biometrics technology that can verify Authentication through extracting the characteristics of the computer to automatically authenticate,it fuses the theory and method of multiple disciplines such as pattern recognition,digital image processing,computer vision,artificial neural network and biometric technologies,face recognition technology research is of great theoretical and practical value.This paper discusses the main face recognition technique of three stages:image preprocessing, feature extraction and face recognition and classification. Firstly,it introduces the face recognition pretreatment geometric normalization method,the image enhancement, the image edge detection and sharpen processing and image binary wait for a method, and feature extraction of principal component analysis algorithm.Then the classifier design the improved method are put forward.(1) Aim to identify of deficiencies of classical BP neural network, we proposes to PSO algorithm of improving learning factor and inertial weights,and the improved PSO(IPSO) use of BP network and the BP network improvement.Using hidden layer number of variable method.Through optimization of BP network,the performance has greatly improved,especially in the accuracy of training speed and discriminated determination;it is more accurate than the traditional methods.(2) In order to further improve the capability of classifier and according to the superiority analysis of flexible neural tree,the paper uses the expression programming algorithm to optimize the flexible neural tree structure,and then puts forward a new method of optimizing the neural tree weights,threshold, learning factor,this method is called selecting particle swarm optimization (RSPSO),finally establishes a complete flexible neural tree model.Through experimental comparison of BP network,IPSO + BP network and FNT three performances,FNT recognition time is shorter,recognition rate is higher,has better classification effect..(3) Simulation experiment By using standard ORL face images and Yale database on experiment,the results shows that the proposed algorithm in recognition effeteness above has greatly improved,especially for face recognition based on neural tree than other methods of experimental results identify effect is better,more efficiency and reliability.