Face Recognition Algorithm Based On Gaussian-Hermite Moments And Sparse Representation

With the rapid development of information science,personal information security has become more and more important.Face recognition is an important means of identifying personal information because of its many advantages such as concealment and uniqueness.In recent years,there have been more and more researches on face recognition.In the research of face recognition,for the reason of the collection equipment and the environment,the occlusion,expression changes,illumination and other disturbances often occur on the face image.It is especially important to design a robust face recognition algorithm.In this paper,the Gaussian-Hermite moments are used to deal with noise and image deformation,and sparse representation classified is used to deal with the problem of illumination and occlusion.The main works as follows:Firstly,we introduce the related background of face recognition and analyzes the current mainstream face recognition algorithm.The Gaussian-Herminte orthogonal moments are introduced to solve the noise interference problem.The non-coefficient Gaussian-Hermite orthogonal moment invariants are derived to deal with problems of face scale transformation,translation and rotation.At the same time,anisotropic Gaussian-Hermite orthogonal moments and tensile invariant moments are proposed.Then this thesis design a new face feature extractor which use the image pyramid to obtain the global and local anisotropic Gaussian-Hermite orthogonal moments and geometric invariant features of the face.The face feature is selected by sctter-ratio.Finally,because the sparse representation classification has strong robustness to illumination changes and occlusion,the feature sparse representation classification is selected as the classifier.At the same time,this thesis introduces how to use truncated Newton interior point method to solve the sparse representation classification model.Therefore,this thesis proposes a new face recognition algorithm based on Gaussian-Hermite moments and sparse representation.The characteristics of three commonly used face databases ORL,FERET and Yale A are analyzed,and a large number of numerical experiments are carried out on the database.The algorithm which is proposed in this thesis is compared with some common algorithms.The result shows that the new algorithm can gain higher recognition rate than other algorithms on the above three databases,and the recognition time can meet the real-time requirements.