Research of Face Recognition Based on Linear Subspace Analysis
|Course||Signal and Information Processing|
|Keywords||Face recognition feature extraction Linear subspace analysis Princial component analysis Fisher linear discriminant analysis|
Face recognition has become a hot researched field in biometric recognition field. Comparing with other technologies of biometric recognition, the technology of face recognition has so many advantages such as non-intrusive, easy-collection and high-accuraccy that it is popular. There are broad applications in the fields of commercial, law enforcement, public security, identity classification, entrance control, video surveillance and so on. Face recognition includes three parts: face detection and localization, feature extraction and classification. This thesis presents the studies of feature extraction in face recognition.Feature extraction is a process which transfers the data from primary spaces into feature space, representing them in a lower dimensional space with less effective characters. Up to now, many methods of feature extraction have been proposed. Among them, the subspace analysis has received extensive attention owing to its appealing properties. Now , the subspace analysis method has been the most popular technology for feature extraction and face recognition.Feature extraction based on linear subspace analysis has advantages, such as high computing efficency and strong geometry feature description ability. Eigenface method based on principal component analysis and the method based on linear discriminant analysis are two succuessful ones.Because of the disturbance coming from environment, one-order feature extraction based on PCA or LDA can hardly get the features exactly standing for identification. By comparing these linear subspace analysis methods in theory and experiment results, This thesis will point out that linear discriminate analysis is more suitable for face recognition than other subspace methods.Fisher linear discriminate analysis has obtained successful applications in face recognition.A new LDA-based face recognition algorithm is proposed in the thesis which is called "weighted null space approach". Due to the fact that Fisher criteria is not directly related to the classification accuracy, we propose a strategy to compute the total between-class scatter matrix and the pooled within-class scatter matrix using some weight functions. A weighted total between-class scatter matrix is constructed in which smaller distances are more heavily weighted than larger distances, because those classes which are clustered together are more likely to be misclassified; a weighted pooled within-class scatter matrix is constructed in which outlier class is more lightly weighed, because one class is well separated from the other classes in the input space , whether the within-class covariance matrix of this class in the new space is compact or not will not have much influence on the classification result. We test this algorithm on ORL and FERET human face databases. Experiment results show that weighted null space method is an excellent face recognition method. It can achieve a high classification accuracy by reserving only a few feature vectors.