Research On Face Recognition Methods Based On Principal Component Analysis

Face recognition is a technology that analyzes the image and extracts effective recognition information to distinguish identity by computer. It has high value of research and good future of application. This thesis mainly focuses on the method for face recognition by principal component analysis (PCA). The main contributions are:(1) A robust eyes detection method called projection based on PCA is proposed. Firstly, this method applies PCA to the facial feature pixels to find the tilted direction for human face. Secondly, it implements gesture compensation. Finally, the eyes are detected by projection. So the proposed method makes the traditional projection method detect the eyes for human face which has gesture changing.(2) Two dimensional principal component analysis(2DPCA)is deeply investigated. Then an adaptively weighted 2DPCA method based on local features is proposed. It combines the structural approach based on geometric feature and the statistical approach based on template through segmentalizing original image into multi-blocks. It intensifies that different features have different contributions to the recognition which are computed by constructing virtual samples. This method not only has the computation simplicity, but also considers the local features information of human face sufficiently. So it possesses the advantages of facial expression, illumination, appearance changing tolerance, and suitability of parallel computation.(3) Two supervised 2DPCA techniques are proposed. One is a weighted 2DPCA method that is put forward by analyzing the rule of 2DPCA from the point of the Euclidean distance. It increases the recognition distances between every two samples belonging to different classes and narrows the recognition distances between every two samples belonging to the same class effectively. The other is to synthesize the between-class covariance and the similarity of the within class to the total covariance. This method's projection vectors can contain the class characters which profits the classification.(4) The advantages and disadvantages of 2DPCA, Alternative 2DPCA, (2D)~2PCA are deeply analyzed. Then a fast (2D)~2PCA method is proposed. It changes the way of computing the projection directions in the second transform of (2D)~2PCA. It can greatly reduce the time of feature extraction and save the storage space.