Face Recognition Based On Improved Center-symmetric Local Binary Pattern

Face recognition technology research and application in recent years has been very broad and deep a biometric technology, it is compared with other biometric technologies, has a non-contact and high security a little, so the identification, human interaction as well as financial security and other fields have a very good prospect. Despite the efforts of people in the face recognition technology becomes more mature, but in the natural environment full of change, face recognition technology is still faced with the expression changes, lighting changes and other factors and challenges.Face recognition process includes face detection, image preprocessing, feature extraction and feature recognition of four main steps, the paper launched a process of recognition of the main research center of symmetry local binary patterns(CS-LBP, center-symmetric local binary pattern) in face recognition, and the paper focuses on the following:(1) Face Recognition Method proposes a single sample filtering and Laplace Center-symmetric local binary patterns(LFCLBP, Laplace filter and center-symmetric local binary pattern) based. First, face image Laplace filter processing; and then filtered to obtain the gradient vector of the image, with the CS-LBP operator to encode gradient amplitude, phase gradient quantization interval to 16 with 4-bit binary encoding, integration into both new LFCLBP features; last block histogram feature, all the subblocks are connected in series to form a human face features, and nearest neighbor classifier to achieve facial feature matching. On the face database experimental results show that the proposed algorithm is effective.(2) This paper proposes a deep learning and center-symmetric local binary patterns(DLCLBP) single sample face recognition method. Firstly, CS-LBP operator face image is encoded to obtain facial features; then will face CS-LBP features as depth belief network(DBN) input, layer by layer training network, formed in the top surface classification; Finally, training good depth belief networks face image samples for identification. Experimental results on YALE face database and ORL database, the results show that the proposed algorithm is effective.(3) Sparse proposed partial phase pattern representation(SRCLPQ) recognition of a single sample. Firstly LPQ phase extraction operator feature face image and its histogram statistics. The histogram feature as inputs, are classified by SQC recognition algorithm. Experimental results on AR face database and FERET face database, the results show that the proposed algorithm is effective.