Face Recognition Based On Boundary Characteristics

Face Recognition is an outstanding ability of Computer Vision. It involves several fields such as Pattern Recognition, Image Process, Psychology and Cognitive Science. It is tie up with both Biometrics-based authentication and Human-Computer apperceive Interaction.Face Recognition began to develop in 1960s and it has gained great achievements. It can be separated into two steps: face detect and localization, feature extraction and recognition. There are three main methods in face detect and localization: detect according to heuristic model, detect according to complexion and detect according to statistics model. As to feature extraction and recognition, it can be divided into two parts because face features can be divided into geometrical features and algebraic features. After extracting geometrical features of sample, we can judge the resemblance between input image and images in database according the distance between them. In order to extract algebraic features, we can do some mathematical transformation for the digital images such as singular value decomposition, K-L transformation. Methods of neural network can also be applied into face recognition.While face detection and localization, we used a method of locating eye features in facial images based on gray-level projection and edge detection. First, the left and right boundaries of face is localized with vertical gray-level projection. Second, face region's horizontal gray-level projection is used localize the top of the head and the middle of the nose. Three, the face structure mode utilized to estimate the positions of eyes, and the edge features are checked to give the proper positions of eyes. Finally we can get standardization face Images with satisfactory effect.We use PCA (the Principal Component Analysis) for feature extraction and recognition, which has been proved to be the top advanced technology of face recognition. First we construct a covariance matrix from sample images, then compute the eigenvalues and corresponding eigenvectors of the covariance matrix, construct a feature matrix with the eigenvectors. Then every image in databasecan be projected into the feature matrix and gain a projection vector, so does the input image. Then we can judge the resemblance between input image with each image in database by computing the distance between their projection vectors. We have used sample images to test the face recognition application based the methods above.