The Method Based On Sparse Representation For Face Recognition

The face recognition technology ushered in its peak period of rapid development, due tothe number of the researchers who devoted themselves to the study of face recognition has arapid development, the computer technology develop rapidly and the requirements for theinformation security is more and more higher in the high-speed information era, the existenceof various factors let the face recognition technology obtained its spring for development. Alot of research institutions, university scholars and graduate students also participate inresearch of face recognition, which has laid a solid foundation for the development of humanface recognition technology. The application of face recognition technology is very wide, ithas a very good application prospect in many fields, such as the public security departmentinvestigation in the case, attendance access control system, bank identification and otherfields, it is an important reason that why the face recognition technology has become aresearch hotspot.Generally, face recognition can be divided into three simple steps: face detection, facialfeature extraction and target classification. First of all, detecting the portion that contains theface from the image, put it as the target image for face recognition. The second step is toextract features of the face image, that is a dimensionality reduction process. Because the faceimage is often a data of relatively high dimension, it contains numerous information, the highdimensional data is not conductive to the classification and recognition process, thecomputational complexity is very greatly. In addition, many information contained in thehigh-dimensional face image are irrelevant for the classification identification, theseinformation were known as redundant information. Feature extraction, also be calleddimensionality reduction, is to remove the redundant information, and extracted the usefulinformation for the classification and recognition, in order to achieve the purpose of reducethe dimension of the face image data, that is more conductive to classification and recognitionfor face image, this is an important step in face recognition. The final step is to use the featureinformation extracted from the face image to classification and identification through theappropriate classification algorithm, then get the class information corresponding to thehuman face image.In this paper, we put forward some improved face recognition method based on analysis of some key problems in face recognition process. After the analysis of sparse representation,we found that the sparse representation method performs robust when the face images sufferfrom illumination, occlusion, facial expression changes and other factors in face recognitionprocess, it can overcome the bad influence taken by these factors and ensure the accuracy ofrecognition in a certain extent. Based on the advantage of sparse representation, this paperpresents the sparse representation-based neighbor graph construction method, and applied it tothe Locality Preserving Projection algorithm and Neighborhood Preserving Embeddingalgorithm, replacing the original neighbor graph construction method based on EuclideanDistance. Then, more groups of experiment were carried on the improved SR-LPP algorithmand SR-NPE algorithm in three well-known face databases: AR face database, Yale facedatabase and ORL face database, the experimental results verified the improved methods offace recognition in this paper can obtain better recognition effect, and achieved greatimprovement in the recognition rate compared with the original LPP algorithm and the NPEalgorithm.