Research On Sparse Representation Image Classification Algorithm

The technology of face recognition plays an important role in the field of information security,more and more stations and airports implement face recognition of self-service ticket channel,not only guarantees in the safety of personnel,but also save a lot of labor,money and time.With the development of face recognition technology,many related algorithms have been proposed.Although these algorithms have achieved good results in various applications and research fields,they still face great challenges in light intensity,posture angle,expression and so on.The main contents and innovations of this master's thesis can be summarized from the following aspects:1.The background and current situation of the research are briefly analyzed.The basic knowledge of sparse representation and the traditional sparse representation classification algorithms are introduced.2.The commonly used algorithm in face recognition is sparse representation.However,in the classification of face images,the algorithm only considers the whole information of the sample,neglects the local structure information of the sample.The local structure of the category contains more discriminant information.Moreover,sparse representation algorithm requires uniform alignment of the test samples and the face images of the training samples.When the posture and angle change,the classification effect is significantly reduced.In this master's thesis,an improved L2 regularization sparse representation algorithm is proposed to solve the problem that the discriminability of sparse representation in face recognition is not obvious,and does not consider the local structure information,combining the whole information of the sample with the local structure information.3.The sparse representation classification algorithm only considers the whole information of the sample,ignoring the influence of the local information of the sample on the image classification.Therefore,when the pose angle,expression and light intensity of the image change,the image classification effect is poor.Based on the traditional algorithm,an improved L2 regularization sparse representation algorithm is proposed.The target function takes into account the overall sparsity of the sample while taking into account the local discriminant of the sample.The sparse coefficient matrix is obtained by the least square method.The residuals of sample reconstruction are calculated and the sparsity measure is used to quantify the sparsity between samples.The target function takes into account the local structure information of the sample,increase the discrimination between the sample classes in maintaining sparse samples at the same time,so the performance of classification is improved.The experimental results show that the method has good classification results on different face databases and extended visual data sets.