Face Recognition Based On Sparse Representation Algorithm

Face recognition belongs to the biometric identification technology,which is unique and portable.However,due to influenced by rich facial expression and external factors such as changing light,occlusion,etc.It is a great challenge to identify with a certain degree of difficulty.Sparse representation theory can effectively overcome the variability of the human face,which achieved widly attention of scholars both at home and abroad.At present,researchers only focuse on the way to design the dictionary or the decomposition algorithm,lead to following problems:(1)The current study has pointed out a variety of sparse decomposition algorithm and dictionary design method,but there are lack of experimental comparison of various methods and performance analysis;(2)Currently sparse representation for face recognition are not separately considered dictionary learning and decomposition method,always directly combinating one dictionary with a specific sparse decomposition algorithm,which made sparse representation can not be best.Aiming at these problems,based on the research results to compare the method of dictionary design and decomposition method separately.Then,according to the construction characteristics of the different dictionary and the decomposition algorithm,throw a large number of experiments to deeply research the sparse representation to get the best ways for face recognition,and providing a reference for researchers to use the sparse representation to solve the other practical problems.The main contents are as follows:(1)Summarized the sparse representation in face recognition from the sparse representation dictionary design and sparse decomposition algorithm.(2)Described the typical redundant dictionary methods and effective sparse decomposition algorithm according to the core research contents,aim at greed tracking algorithm and relaxation optimization algorithm,through experiments with one-dimension signals and two-dimensional image signal to compare and analysis the different performance of OMP,TNIPM,Homotopy,FISTA,PALM and DALM,which has a certain guiding significance.(3)Studied the sparse representation classification algorithm based on the analysis dictionary for face recognition.Basis on SRC algorithm,Extended Sparse Representation-based Classifier and Superposed Sparse Representation-based Classifier algorithm,through experiments to assess the accuracy of SRC algorithms and ESRC,SSRC algorithm in face recognition.Besides,this paper has used a variety of sparse decomposition algorithm to improve the SSRC algorithm,proposed TSSRC,PSSRC,DSSRC,FSSRC algorithm.Single sample face images,FERET and AR face database were used to compare the algorithm performance and finally the better sparse representation was given to solve practical problems.(4)Studied the sparse representation classification algorithm based on the learning dictionary for face recognition.Used classic K-SVD algorithm,Discriminative K-SVD,Label Consistent K-SVD,Fisher Discrimination Dictionary Learning method to assess the performance in face recognition.And improved FDDL dictionary learning process with a variety of sparse decomposition algorithm,proposed FDHDL、FDTDL、FDPDL、FDDDL、FDFDL algorithm to get FDFDL was the best one in both the accuracy and computational efficiency.(5)Face recognition via sparse representation system was build.Choose the Yale Face database and lab collection library to test the system.Experimental results showed the improved SSRC and FDDL algorithm in face recognition not only improved the recognition accuracy,but also effectively improved the efficiency.