Research On Face Recognition Based On LBP Operator

Automatic Face Recognition (AFR) holds an important position in various biometrics techniques for its superiority. With more than 30 years'development, AFR has made great achievements. The state-of-the-art AFR system can perform identification successfully under well-controlled environment, and many commercial AFR systems have appeared. However, due to the complexity and uncertainty of face recognition, there are still many key problems to be resolved for further application of AFR. Feature extraction is the crux of face recognition problem, which directly related to the selection of the classification algorithm and the accuracy of the system.The local binary pattern (LBP) operator is defined as a gray-scale invariant texture measure, derived from a general definition of texture in a local neighborhood. It was first introduced as a complementary measure for local image contrast. Recently, the LBP has been successfully applied to face recognition as texture descriptor and excellent result has achieved. However, there are still many limitations in the basic LBP operator and the LBP-based face recognition algorithm. To resolve these problems, the dissertation is devoted to the investigation on LBP and its application in face recognition. The main contributions of the dissertation are as following:1. The dissertation investigated the dimension reduction of LBP patterns. To overcome the limitation of the uniform LBP patterns, a new LBP sub-patterns was developed to reduce the dimension of the descriptor by utilizing PCA. In comparison with the LBP uniform patters, the proposed LBP sub-patterns can extract the most typical sub-patterns features, and reduce the dimension flexibly, conveniently and effectively. Furthermore, the LBP sub-patterns can help to eliminate the noise for accurate feature extraction.2. The dissertation investigated multi-scale LBP and proposed a new LBP pyramid method. This method firstly constructs a multi-scale space of the image by utilizing multi-scale filtering. Then the LBP operator is applied to each image in the multi-scale space. In comparison with existing multi-scale LBP methods, the proposed LBP pyramid method can extract abundant features accurately and effectively with lower computational cost.3. A novel algorithm for face description and recognition based on multi-scale LBP sub-pattern feature is proposed in this dissertation. The proposed algorithm combines the local description ability of LBP and the global description ability of PCA, which can effectively extract both the structural features and statistical features for face recognition. Furthermore, the dimension of the extracted features is lower, which satisfies the requirement of real-time application. Experimental results shown that the multi-scale LBP sub-pattern feature is highly discriminable with good performance in face feature expression and is robust to illumination, face expression and position variations.The dissertation mainly investigates the LBP operator and its application in face recognition, brings forth new ideas on LBP dimension reduction, multi-scale LBP feature extraction, and LBP-based face recognition. It is of active significance to promote the development and application of automatic face recognition technology.