Face Recognition And Facial Feature Localization Based On Non-Local Binary Pattern

Face recognition technology is a challenging task with very important value. It is an active research topic within research communities like Pattern Recognition, Computer Vision, etc. As an effective operator for face representation, Local Binary Patterns has received more and more attention from researchers. However, because of its deficiencies, LBP needs to be adapted to solve the problem of face recognition.In this thesis, we propose a new operator named Non-Local Binary Patterns. Compared to original LBP operator, the new operator has the following advantages: 1) The NLBP operator can implement multi-scale image analysis; 2) The NLBP operator is capable of enhancing and augmenting those salient facial features, which are important for classifying face images; 3) The NLBP operator can increase the robustness of face representation against occlusion and misalignment.Facial features localization is one core problem in the field of face recognition. This thesis employs our proposed NLBP operator to design two facial feature localization methods which can be applied to fulfill different tasks in Face Recognition System:Facial feature regions detection method based on NLBP operator; NLBP operator with coarse scale has the effect of augmenting the regions of facial features and excluding most noises from irrelevant regions, we make use of this and design a method which is able to implement fast and accurate detection of regions of facial features. This method can be used as a pre-processing technology for many special face recognition methodsAccurate eye detection method based on NLBP operator; The proposed algorithm consists of two main stages. Rough eye regions are first detected and then the eye pair will be pinpointed through the way of evaluating the similarity of each eye-pair candidate within the rough regions. This method exhibits reliable performance in experiments. It can be employed for the task of aligning face images or predicting face pose.We combine the use of Face Detection and our two localization techniques in the whole still and video images. The application of the information extracted by these techniques in Face Recognition System is also demonstrated in the thesis.