Research Of Robust Face Recognition Based On Block SRC

In recent years,with the development of computer technology and pattern recognition,face recognition technology has been applied in various aspects,such as face verification,access control,security,human-computer interaction,smart card,law enforcement,multimedia management,and monitoring.Due to the increasing demand of the application,the research of the face recognition technology has been achieved better results under the controllable conditions.At the same time,there also have some problems needs to be solved : partial information loss caused by occlusion,deformation of face structure caused by the change of posture and expression,the change of face texture caused by age,the size of the face database,etc.Aiming at the problem of occlusion and illumination change in face recognition,we improved the sparse representation classification algorithm to improve the rate of face recognition under the condition of illumination and occlusion.The main work is as follows:(1)According to the uncertainty characteristics of face cover parts,a block SRC algorithm is proposed in this paper,which combines the local and global features of the face,and has robustness to face recognition under occlusion conditions.Firstly,all the sample images are segmented in a certain way,and then each sub-block is classified according to the sparse representation.Finally,the system based on the classification criterion of fusion global and partial feature fusion to complete face recognition.Compared with traditional classification methods,the algorithm presented in this paper combines the Borda voting method and residual ratio to determine the classification,not only overcome the problem of overgeneralization in local discriminant algorithm,but also overcome the problem of disagreement inoverall method.By using the standard AR and Extended B Yale face databasefor experiments,the simulation results show that the algorithm can significantly improve the rate of face recognition under occlusion conditions.(2)Aiming at the problem of illumination change on face images,this paper presents a robust algorithm based on feature blocks-weighted sparse representation classification.Firstly,for all images based on log-DCT transformation to realize the illumination normalization.Then,the image of the pre-processed image is segmented based on the features(such as forehead,eye,nose,mouth and so on),and a weighting matrix W is introduced.The control experiment is carried out with each feature sub-block weight as a single variable to test the contribution of different feature sub-blocks.According to the test results,different weights are given for each feature sub block.The log–DCTcan compensate the uneven illumination of the face image,and the weighted block can better express the face information by setting different weights to different parts of the face.The recognition rate of the proposed algorithm on Yale Bface database and Extended YaleB face database was 97.45% and 98.10%,respectively.The experimental results verify the effectiveness of the proposed algorithm in face recognition under complex illumination conditions.(3)Based on the proposed feature blocks-weighted SRC algorithm,an identity authentication system is implemented to use the face as a network application login.Apply the identity authentication system as network(micro-blog,BBS,etc.)of cloud services,allow the network application of authorized to log in to face identification authentication.The use of human face instead of the traditional digital as a network application login password,not only eliminates the cumbersome password memory,but also effective protection of user privacy.At the same time,through the performance test of the system,wecan also find out the shortcomings of the algorithm for subsequent improvement.