Face Recognition Base On Adaptive Soft Histogram Local Binary Pattern

As an important biometric technique, face recognition has become a hot topic in image processing and pattern recognition during recent years, and is widely used in many areas such as public security, authentication and public administration. Feature extraction is the first and key step of face recognition, and impacts intensively on the effectiveness of a face recognition system.As an effective facial descriptor, local binary patterns (LBP) has obtained much attention these years. Soft histogram local binary patterns (SLBP) descriptor enhances the robustness of the LBP descriptor, but it has some inherent defects. Based on the analysis and comparison of LBP, SLBP, we propose an adaptive soft histogram local binary patterns (ASLBP) algorithm. The main research work and innovative contributions are as follows:1. Based on SLBP, we have proposed an adaptive soft histogram local binary patterns (ASLBP) algorithm. Firstly, we have proposed an energy function of training samples. Then the gradient descent method is utilized to obtain the local minima of this energy function. Finally, the best fuzzy membership function for SLBP is obtained by the local minima of the energy function. By using the Fisher discriminant ratio (FDR), we have changed the problem of selecting two fuzzy membership functions of SLBP to an optimization problem, with a nonlinear objective function and several linear constraints. The face recognition system could adaptively refine the fuzzy membership function with the aim of improving recognition accuracy.2We have also applied the same learning procedure to the local ternary patterns (LTP) and obtain the adaptive soft histogram local ternary patterns (ASLTP) algorithm3. We have designed experiments to compare the effectiveness of these algorithms on the CMU-PIE and Extended-YaleB face databases. The results indicate that the proposed ASLBP algorithm outperformes LBP and SLBP. The performance of ASLTP is aslo superior to LTP. 4. Based on the above research work, we have developed a real-time face recognition system, which includes image acquisition, image preprocessing and face recognition technology.