Face Descriptor Based On LBP Sampling Learning

In the past few decades,face recognition has attracted much attention in the field of pattern recognition and computer vision,due to its huge application prospect and theoretical significance.Local binary patterns(LBP)is an efficient local feature and has been widely used in many face recognition systems due to its strong discriminative power and excellent robustness.However,most existing LBP-like face descriptors are hand-crafted,which require strong prior knowledge to engineer them by hand.The present work mainly studied on the learning-based LBP face descriptor,and proposed a learning-based hard sampling pattern as well as two new face descriptors.The main contributions of this paper are summarized below:1.First,we studied the discriminant local binary patterns(DLBP).To overcome the drawbacks of its complexity and difficulty to solve,we proposed the PDV-DLBP hard sampling pattern based on this.It could measure the distance of two images using pixel difference vectors(PDV).Moreover,we established the objective function based on Fisher's linear discriminant.It should be noticed that we converted the solving problem of PDV-DLBP into a standard 0-1 fractional programming problem,which had lower time complexity than DLBP.Compared with DLBP and PDV-DLBP algorithm,the results on FERET and AR face database indicated that the developed model was more rapid,and had higher recognition rate.2.Considering the consistency of PDV-DLBP hard sampling learning and Image Filter Learning based LBP(IFL-LBP)in objective function,a more discriminant IFL-DLBP face descriptor was constructed by combining the two algorithms.It could learn the image filter and hard sampling pattern at the same time.We discovered that the sampling neighborhood of Discriminant Face Descriptor(DFD)had a great influence on the final recognition performance through experiments.Thus,we introduced the PDV-DLBP hard sampling pattern into DFD.It solved the problem of descriptor neighborhood selection and then the Filter Sampling-Local Binary Patterns(FS-DLBP)face descriptor was obtained.In particular,FS-DLBP is more discriminative compared with other descriptors.Finally,through detailed experiments on FERET and AR face database,the results verified the high recognition performance of the proposed IFL-DLBP and FS-DLBP face descriptors,as well as the robustness to variations of expression,illumination and other interference factors.