Design And Research Of Face Recognition System Based On PCA

As an important branch of biometric recognition,face recognition is characteristic of security,reliability,intuitiveness and noncontact,which is widely used in many fields,such as identity verification,public security investigation,video surveillance,national defense security.Therefore,it is of great value and realistic significance to research and design a fast and accurate face recognition system.The research of this thesis is mainly about the face recognition system based on the Principal Component Analysis(PCA)method.The main contents of this paper are as follows:(1)Image preprocess.The images are normalized by image grayscale transformation,histogram equalization,smoothing,geometric correction and some other methods in order to meet the requirements of the recognition system.(2)Image feature extraction.The eigenvalue and eigenvector of face images were obtained through K-L transform and SVD theorem,achieving the purpose of image dimensionality reduction.Thereby,the amount of calculation is reduced greatly and the recognition speed is improved significantly.(3)Image feature recognition.The traditional PCA-based face recognition method accomplishes the face recognition function through using the distance function,but the recognition accuracy rate of this method is low.Therefore,we came up with a solution of Support Vector Machine(SVM)classifier.The improved face recognition system obtained high recognition accuracy rate and good practicability,which is obviously better than the traditional PCA-based face recognition method on recognition accuracy rate.(4)System design and simulation implementation.First,we designed a face recognition system based on PCA+SVM in detail and performed a large number of simulation experiments on Matlab software by using ORL face database.Then,by changing the feature face space dimension and the training set face number,we obtained several sets of experimental results.Finally,through the detailed analysis of the experimental results,we got the best parameter combination of the system.