Research Of Face Recognition Algorithm Based On Improved LBP

Face recognition technology is a very important area of pattern recognition, a lot of researchers have done investigation on it in recent years.Face recognition has been used in many places, such as airport security, smart guard, police surveillance, intelligent video surveillance, immigration control, driver license verification and so on. In order to obtain reliable recognition rate, a lot of face recognition methods has been obtained, feature extraction technology is extremely important, because the extracted feature of face images will directly affect the the recognition rate.The local binary pattern operator is a simple and effective feature extraction operator, which is used widely. In this thesis, local binary pattern operator and principal component analysis are used to exploration and research.Firstly, the method of multi-scale local binary patterns and principal component analysis is proposed, the feature of a face image is extracted by local binary patterns with a different square window, then the local binary patterns feature of each scale are linked separately, all features obtained at each scale are linked together as the multi-scale local binary patterns of the face image. The principal component analysis is used to reduce the dimension of features,Euclidean distance is used to deterrmine the category of test image. This chapter introduces the multi-scale LBP firstly, and then introduces the process of extracting feature, the recognition rate of proposed algorithm. Finally, Yale-B face database is used to simulation experiment,the results shows that the recognition rate of this algorithm is comparable to multi-scale local binary patterns, but the time spent is less than multi-scale local binary patterns. On the while, the recognition performance of the algorithm is good.Secondly,the method of improved local binary patterns and principal component analysis for face recognition is proposed, the traditional local binary pattern uses only the symbol component, while the improved local binary patterns which uses symbols component and amplitude components to describe the facial feature better. Firstly, the method encodes the symbol components of gray difference between center pixel and its neighbors’, reflecting local structures of the face; Secondly, it uses the amplitude components of gray difference between center pixel and its neighbors’ as the complement of a local binary pattern; Finally, it uses principal component analysis to reduce the dimension of feature descriptor and enhances the ability to discriminate. Numerical experimental results on ORL face database shows that the proposed method can increases the recognition rate of face images markedly.