The Technology Of Face Recognition Based On Improved LBP And AdaBoost

Face recognition technology has been used in the fields of military, finance, public security, video monitoring system, etc. Feature extraction is the core of face recognition technology and it will directly affect the recognition performace. In recent years, the local binary pattern(LBP) and its extensions have been widely used in face recognition for its simplicity and effectiveness. In this paper, we also concentrated on LBP and its extensions.1. In view of the importance of local binary pattern(LBP) in image processing and computer vision, the current extensions of LBP in different application fields are reviewed systematically in the paper. Firstly, the principle of the basic LBP operator is briefly discussed. Then, the extended methods of the LBP are conducted and summarized from five aspects including the neighborhood topological structure, noise resistant, dimension reduction, coding methods and rotation invariance respectively. Finally, the mutual relations among the five aspects and the problems of each type are analyzed in detail and the future for the LBP extensions is pointed out.2. A new method is proposed in the paper bansed on EQP(Elongated Quinary Pattern) for face recognition. Though EQP is effective for face recognition, it is sensitive to the changes of image gray by the global threshold. On the other hand, different sub-blocks are treated equally in face recognition based on EQP. A new improved method, called robust EQP(REQP), is proposed in the paper. Firstly, an adaptive threshold is determined to meet the robustness of image gray change. Furthermore, each sub-block is weighted by its structural contrast to emphasize the different roles of different sub-blocks. Experimental results obtained from two widely used face image databases, YALE and ORL, demonstrated that the proposed method can greatly improve the performance of the traditional EQP on face recognition.3. Though REQP is effective for face recognition, the dimensionality of the extracted histogram is high. Two kinds of feature selection methods are discussed to reduce the redundant information for dimension reduction. After that, Adaboost is adopted as the the classifier for face recognition together with the new feature. The experimental results tested on YALE and ORL image databased show that the new method can improve the recognition rate greatly than REQP.