Nuclear Norm Based Two-dimensional Maximum Margin Criterion

The face recognition technology has become a hot research topic nowadays.It has a very important significance in both theoretical and practical application.In recent years,with the rapid development of the Internet,the updating of the communication technology,the face recognition technology has been a great development,but there are still some problems.For example,the rate of recognition will be reduced when the image is disturbed by large noise,light,the expression of face and the changing of shooting angle.So the research in this area needs to be deepened continuously.Based on the face recognition in this paper,the writer briefly introduced the related content of the face recognition technology,including the meaning and the background of the research,the achievements and the research progress nowadays both in domestic and overseas.In the face recognition,the feature extraction is a key step,in this paper several typical algorithms are described,including one-dimensional subspace analysis method,two-dimensional subspace analysis method and the subspace analysis method based on nuclear norm.The two-dimensional maximum margin criterion has attracted wide attention in the face recognition,it is a supervised feature method based on two-dimensional,which can better preserve the structure and category information of the sample image.And this algorithm effectively overcomes the "small sample" problem in the linear discriminant analysis.However,in practical applications,the face image is often disturbed by sparse large noise and light and so on.It is difficult to overcome the impact of these environment by the two-dimensional maximum margin criterion,which can lead to wrong classification.In order to solve the problem,the writer proposes a new method in this paper—— nuclear norm based two-dimensional maximum margin criterion.It uses the nuclear norm metric instead of the Frobenius norm,and it takes the alternately iterative algorithm to obtain the two optimal projection matrices,then the writer uses matlab software to perform face recognition experiments on both ORL and Yale database,the experimental results show: the nuclear norm metric is not easily affected by the sparse large noise and light in the calculation,it can be more accurately classified in the face recognition.