Based On Singular Value Decomposition And Sparse Representation Of Face Recognition

The first step to recognize a person is to recognize his face which contains the information about his identity, gender, age, etc. With the rapid development of computer technology and graphics technology, the computer pattern recognition problem which based on the face image has become a hot issue in recent years. Compared with other biometric identification technology, iris and finger prints for example, face recognition technology has the advantages of naturalness and imperceptibility. Therefore, face recognition has a broad prospect of application in identity recognition, camera surveillance and so on.People have researched for over decades of time on face recognition, although some headway has been made both in theory and practical, the existing face recognition method is limited in applications. There’s many elements make face recognition inaccurate, such as the selection of face feature, the quantities of training samples, the influence of light and shade, the change of posture, etc. Compared with other face recognition method, the one based on sparse representation has a better recognition rate and robustness, but in the situation of less training samples and characteristic vector dimension, the recognition rate is not ideal. Based on this kind of situation, a method of face recognition based on singular value decomposition and sparse representation is brought in this paper.For the situation of less training samples and characteristic vector dimension, the recognition rate of face recognition via sparse representation is not ideal, an improved face recognition method is brought forward, which based on singular value decomposition and sparse representation. First of all, an extraction method based both on global and local feature is brought forward while its algorithm principle is elaborated. This method is piecemeal the face image to get local characteristics in each block by using singular value decomposition, then get global characteristics by using singular value decomposition on global image. Secondly, a specific calculation process of face recognition based singular value decomposition and sparse representation is introduced. Finally, a specific method of solving sparse solution is introduced. The choices for the singular values of global image and sub image has be discussed. By doing the experiment on ORL face database and analyzing the results which come from the method that brought forward in this paper between the one come from the method that literature [12] used, proves the method in this paper effective and stable.