Research On Face Recognition Algorithm Based On Sparse Representation

Face recognition technology is considered to be the most effective biometric recognition system because it requires the least human-computer interaction.Therefore,a large number of experts and scholars have been attracted to the research of face recognition,and many related algorithms are proposed,some of which have been applied to real life and achieved good results.However,there are still some unresolved problems in the actual operation process.For example,face image acquisition is easily interfered by other conditions,resulting in insufficient samples.In addition,the conventional sparse representation method does not take into account nonlinear information in the face image.These will affect the efficiency of face recognition,so this paper designed two improved algorithms for these practical difficulties.1.This paper designs a discriminative face recognition method based on kernel sparse representation.It consists of three main parts:l2 regularization,kernel trick and discriminative face representation.First of all,the method is based on the l2 regularization framework,compared with the traditional method of solving the representation coefficient based on l1 regularization,the operation efficiency of the algorithm is guaranteed.Secondly,the general face recognition methods ignore the internal relationships of the image,such as nonlinear information,in the classification process of the original sample space.This article uses the kernel function to solve this problem.Finally,by setting a special objective function,the method can effectively enhance the difference between different classes of face samples,so that the discriminative representation can be generated for each test face image,which further improves the recognition accuracy of the algorithm.Experiments on public face sets show that the method has good characteristics.2.In field of face recognition,a key issue is whether there is a sufficient number of face training samples with valid information.However,due to the complexity of human face images,face recognition is easy to be affected by the external environment such as light intensity,expression,hair style and occlusion.Therefore,it is difficult to obtain enough effective samples in real operation.In this paper,we design a new algorithm that generates virtual images by utilizing the singular value decomposition method.The virtual images not only increase the number of samples of the dataset,but also adapt well to the test image.In addition,we use the weighted score fusion scheme to calculate the ultimate results,which can make better use of the data from different sources,including the original images and virtual images.The proposed algorithm is executed on several face sets,and the final experimental results confirm that the second algorithm proposed in this paper can obtain satisfactory performance.