Single-Sample Face Recognition Based On Complex Conditions

At present,biometrics technology has been widely used to supervise law enforcement systems with the development of science and technology and the advancement of equipment.As the most widely used face recognition technology in biometric technology,it has the advantages of non-mandatory,multi-popular,friendly and fast.Face recognition technology has become one of the most popular and widely used technologies.However,in some practical application scenarios,only one training sample of the face recognition system,such as passport verification,terrorist tracking and video surveillance,can only collect a standard face image on the test individual ID.As a training sample,it is difficult to extract the discriminative features that can accurately express the information changes in the face class because only a single training sample,and many traditional face recognition methods based on multi-sample training will no longer be available.Moreover,in the actual operation of the face recognition system,the facial features of the subject may be interfered by many uncontrollable factors,such as changes in expressions and gestures,occlusion of glasses and accessories,and strong light angles,etc.,in a single sample person.Under the premise of face recognition,these interference factors can seriously lead to the unsatisfactory effect of the face recognition system.Based on the above two aspects,this paper proposes innovations in the expansion and classification of single sample sets to further improve the performance and robustness of single-sample face recognition.In this paper,a general sample set expansion training sample based on information with changing characteristics is proposed.The SRC model is combined with the local segmentation idea to calculate the most sparse matrix coefficients,thereby improving the recognition effect and reducing the computational complexity.The specific work is as follows:(1)Sparse Representation with Extended Generic Set(SREGS)one-sample face recognition method is proposed.The method learns the face variation feature based on a universal face sample set containing V face change information,and reconstructs the face training sample set by mathematical model superposition.Furthermore,the number of face image samples and the feature dimension of each object in the training set are improved.Finally,the dictionary of the face is expanded and identified by means of the sparse representation frame model.Experiments were carried out on AR,Extended Yale B,and LFW standard face databases,and compared with some classic algorithms such as SRC,ESRC,SVDL,and DMMA.The experimental results show that the proposed method not only guarantees the completeness of the training samples,but also the number of face recognition in the single-sample face recognition under complex changing conditions.Also improved.The experimental results verify that the innovation has good robustness and implementability in the application of single-sample face recognition.(2)Sparse Representation based on Local Spatial Fusion(SRLSF)single-sample face recognition is proposed.Since the SRC model requires that the dictionary of sparse representations must be over-complete(that is,the number of samples is sufficiently larger than the sample dimension),although the training sample set can be extended to an over-complete condition by a universal sample set containing V face-changing features.However,constructing a universal face set containing various complicated conditions,the operation complexity violates the convenience and convenience of face recognition.In this paper,the expanded face samples and test samples are divided into non-overlapping and equal-sized local blocks,and all local sub-blocks are located in the same linear space.The sub-blocks are generated into columns by the vector and then input into the SRC model.Finally,each local area SRC minimum residuals are used as the majority voting strategy for identification.Experiments were carried out on the ORL,FERET,and CAS-PEAL standard face databases,and compared with some classical algorithms such as ESRC,SVDL,FLDA,PCRC,PSRC,LGR,FLDA-SVD,and the experimental results verified the algorithm proposed in this chapter.Innovation is effective in testing faces under the influence of complex conditions,and the recognition rate has been greatly improved.