Face Recognition Research Based On Collaborative Representation

With the rapid development of society,the problem of information security has been paid more and more attention.How to carry out accurate and rapid identification has become an important problem to be solved urgently.Face recognition technology uses the unique facial features of each person to identify,compared with other biological identification technology,face recognition has some advantages like simply collection,non-invasive,consistent with human cognitive habits and so on.Face recognition technology is a hotspot in the field of pattern recognition and artificial intelligence,and it shows great application prospect in public security and economic finance.Although the face recognition technology has made many breakthroughs,in the uncontrolled conditions,the recognition effect is not ideal,it still has a high degree of challenge.This paper first introduces the background,significance,research status and general algorithm flow of face recognition,then studies the feature extraction and classification method in face recognition.In this paper,the Sparse Representation based Classfication(SRC)theory is studied as a focal point.To solve the problem of its high computational complexity,we introduce the Collaborative Representation Classfication(CRC),and the computational complexity is reduced by using the l2 norm instead of the l1 norm.Through the comparison of these two algorithms,it is found that the target of SRC and CRC algorithm is to use the training samples to achieve the optimal reconstruction of the test samples,and then determine the category of the test samples by the coding coefficients.The computational complexity of the collaborative representation algorithm is low and more real-time,so the improved algorithm of collaborative representation is more suitable for practical application.Therefore,this article has made a series of improvements to the shortcomings of the collaborative representation method.The main innovations of this paper are as follows:(1)Due to the unavoidable noise error in the training samples,errors occur when the collaborative representation algorithm reconstructs the test samples.In view of this shortcoming,this paper proposes a method to deal with face images with low rank matrix recovery(LR)algorithm to eliminate the noise in face images and to highlight the common features of the same human faces in order to reconstruct the test sample better.In the calculation of the low rank approximation matrix,the inaccurate Lagrangian multiplier method(IALM)is used instead of the exact Lagrangian multiplier method(EALM)to reduce the computational difficulty and improve the operation efficiency,make the algorithm more practical.(2)The collaborative representation algorithm ignores the importance of the neighbor samples in reconstructing the test samples.In response to this shortcoming,local consistency is introduced into the coding scheme so that the coding coefficients of similar samples(neighboring samples)are as similar as possible.And low-rank recovery can effectively remove the original picture of the noise and error,highlighting the common characteristics of the same face pictures,making the same face pictures as close as possible.Combining these two methods,a new face recognition algorithm based on low rank recovery and locality-constrained collaborative representation(LR-LCCR)is proposed,which can make the coding coefficients of the same face images similar and enhance robustness.Experiments on the AR and Extend Yale B face databases demonstrate the effectiveness of the algorithm.(3)Sparse representation and collaborative representation usually use the image's global characteristics to form the training library dictionary.Since the global feature description of the face image is not sufficient,it is improved for the dictionary of collaborative representation.In this paper,an effective method for extracting local features of face images is presented,as Patterns of Oriented Edge Magnitude(POEM).The POEM features are used to constitute the dictionary of collaborative representation and an collaborative representation algorithm based on POEM feature(POEM-CRC)is proposed.Experiments on AR and FERET face databases demonstrate that POEM-CRC algorithm performs better than other similar algorithms.