Sparse Representation And Dictionary Learning Based Face Recognition

In the field of pattern recognition,face recognition is known to be a hot research field.Face recognition used in many areas,such as public security systems,high-speed rail stations,running a red light check.Compared with traditional fingerprinting,face recognition has great advantages.Fingerprints can not be recognized when the air is dry,which has caused great troubles for some important applications.With the development of face recognition technology,This paper analyze and summarize the current classification algorithms based on sparse representation,and propose the following three improved classification and recognition algorithms:1.A structured sparse representation of face recognition algorithm based on CLBP feature extraction is proposed.CLBP algorithm is an improvement of LBP algorithm.In the CLBP feature extraction.The CLBP feature extraction can make better use of the local features of the face image.Then,the extracted CLBP histogram features are input into the structural sparse representation(SSRC)classifier.This algorithm make full use of SSRC algorithm training dictionary block structure,effectively improve the face recognition effect.2.Based on the above,we propose a kernel sparse representation algorithm(DG-KSRC)based on the new Gabor feature recombination method.In this method,40 Gabor feature sets are grouped together in 5 scales and 8 directions,and then input into our KSRC model based on the coordinate descent.The algorithm can take full account of the independence of Gabor face features at different scales and directions.The algorithm has better robustness to changes such as lighting,expression and occlusion.3.A MOD dictionary learning algorithm based on semi-supervised competition aggregation is proposed.We propose a MOD dictionary learning algorithm based on semi-supervisor competition and aggregation(SCA-MOD),which adds a clustering algorithm to the classic MOD algorithm.In the dictionary learning stage,we first use the semi-supervised competition aggregation algorithm to remove redundant dictionaries Atom,and then use the MOD algorithm to learn in a dictionary without discrimination.The final classification dictionary is better.The feasibility and validity of the proposed algorithm are verified by conduct comparative experiments on the three common face database AR,ORL and LFW...