Reasearch Of Face Recognition Based On Sparse Representation

Face recognition belongs to the technology of Biometric identification.It is an important subject in the field of computer vision.Through the using of the unique character of individual biometrics,it realizes the individual identification.Because the recognition technology is without contacting that makes its application more hygienic and efficient,then it is widely used in intelligent security?Entrance guard?computer vision and other fields and is the important Research direction of artificial intelligence.So we start the research of the problem in this paper.First for the difficulty about satisfying the richness of face gesture in the training samples set and its affect on the rate of face recognition,the paper proposes a method that generates virtual gesture images for recognition,which uses the redundancy information between two different facial images to establish the correspondence relation between the sub-blocks from two images in the vertical direction.Then basing the correspondence between the two face images,it can make regional telescopic to produce a new image using the average of two new.We add the virtual posture images to the training images set so that to increase the diversity and linear representation ability of training set with multiple face pose and to improve recognition rate as a result.Second for the problem of image registration in the recognition,we use an automated matching method based on sparse representation to get the best registration position between the test image and each subject of the train set by solving an optimization problem,resulting in the registration error of each subject.After analyzing the distribution characteristics of the registration errors,the paper proposes a method that sets the class parameter adaptively and then adaptively makes the size of sparse representation dictionary,at last resulting in improving the recognition accuracy rate of sparse representation classifier.Based on the research above,the paper performs experiments on the international standard test database FETET.In the tests,this paper selects partial sample in each subject of the database as training samples and the remaining as the test samples.First we add the virtual face images generated by virtual face generation method in this paper to the existing training samples and then train the face recognition classifier to test its ability of identified.The result shows that with the virtual samples it has a higher recognition rate.Second we choose the same training and test samples in the database to compare recognition rate between the method of adaptive parameters sparse representation recognition and the original method,the test results show that this method has a higher recognition rate thus verifying the validity of this method.