Face Recognition Algorithm Based On Multi-module And Sparse Recognition

In recent years, face recognition has been a popular research area in biometric. In actual application,however, due to the face image changes existing in the expression, noise pollution, illumination and occlusion makes face recognition facing with challenges.Based on the research of many scientists, this paper aims at solving these problems for further research on the basis of the sparse representation classification.Firstly, classic module classification methods usually use consistent judgment model.They treat each module as the same and neglect the data distribution differences caused by the different content. In this article, we propose a multi-module classifier based on the sparse representation to optimize the method. First block the training sample, testing sample and validation sample. Then using three different methods classify each module in the validation sample respectively. By comparing the judgment results of each module, it can be obtained the best method. Then select the best optimization classification model to classify each module. At last, count the result of each module, it is concluded that the test images of the final judgment.Secondly, for most face recognition algorithms usually extract linear face feature space. They ignore the nonlinear structure of face image. Apart from this, face recognition algorithm based on the whole face and ignore the local characteristics, thus affecting facial recognition robustness. This article proposes the kernel sparse representation classification algorithm based on blocks. First block the face image, and then map each module to the kernel space respectively. Then use kernel sparse representation to classify each module,finally analyze the discriminant result and get the final category.Finally, in reality, face image often has a illumination, occlusion or noise pollution problem. This paper proposes tow-stage sparse representation based on the low-rank. First use low-rank to deal with the face image to get low rank structure. And then learn a robust metric and map the cleaning samples to the new space. Then use the nearest neighbor method select the most similar images from the training samples, reconstitution the training dictionary. Finally use sparse representation classification to get the result.