The Research And Application On Sparse-representation Based Face Recognition Methods

In the recent years,due to the rapid development of visual sensors and the mode recog-nition technology,face recognition has been a popular area and important technology in biological information recognition field.At the moment,in the study of face recognition,both of theoretical studies and applications have gained great achievement,especially for the typical and static face photos,and in this kind of databases,mainstream face recognition methods are able to obtain very high accuracies.However,in the practical applications,the final recognition result often be interfered by a number of factors including lighting condi-tions,postures,obstructions,the lack of samples.Therefore,how to improve the accuracy of face recognition become the research priorities of scientists,and also the main direction of this paper.This paper simply analyses the fundamental rationale of spare representation-based face recognition methods and virtual sample generation methods,and according the char-acters of spare representation classification proposes a group Group Representation-based Classification method,moreover this proposes a virtual samples based two phrases face recognition method combining the virtual samples generations technology.Specifically,the main research works and contributions are summarized as follows.From the perspective of the relationship between image dimensions and training sam-ples' quantity in spare representation classification,this paper build a new face recogni-tion method called Group Representation-based Classification.Due to alleviate the impact of over-complete dictionary with excessive samples,this method divides the training sam-ples to several groups,and thereby obtaining higher accuracies,additionally this method is able to be applied to other spare representation-based classification algorithms.In order to demonstrate the effectiveness of this method,a large number of face recognition experi-ments have been run and the results on different face databases show that this method has better classification results than those traditional approaches.In order to obtain a better accuracy,this paper consequently proposes a virtual samples based two phrases face recognition method which is built on the Group Representation-based Classification and takes the advantage of virtual sample generation technology,be-sides this method also use the HOG to reduce the dimension of sample images.Although this method obtains excellent recognition results with a low calculative complexity,the se-lecting to optimal number of groups remains a problem.In this paper,all optimal G values are confirmed by the experimental results,without figuring out the rationale,which is the next step of our study.Finally,experiments proof that this method has a better accuracy than traditional approaches.