A Novel Sparse Representation Method Based On Virtual Samples For Face Recognition

Face recognition is an important technology of modern biological information recognition.And when compared with other biometric identification technology,it has more advantages such as intuition,simplicity,extendibility and so on.As for it,face recognition is widely studied and applied in recent years.So far,large numbers of face recognition mathematics have been proposed and even some of them are perform very well.But for face recognition can be greatly interfered by a lot of factors such as posture,facial expressions,lighting conditions,shelter materials and lack of training samples and so on.So it will bring big disturbances to face recognition,and finally made the face recognition methods cannot recognize face images accurately and the authentication work may be not so successfully.Therefore,how to reduce the recognition error rate is the direction of the development of face recognition,and also the original intention of this paper.The method of sparse representation has low recognition error rate and strong robustness.This paper focuses on its application in the field of face recognition,first elaborated the sparse representation method base on virtual sample,and verified the feasibility in face recognition through the experiments.The main works of this paper include:In order to solve the problem of insufficient capacity of training sample,which caused by variant face expression and changing posture,this paper puts forward the improved sparse representation method based on virtual sample.This method uses facial symmetry to generate virtual samples,and uses these virtual samples to simulate the possible view of face image under the condition of variant face expression and posture change.Because these virtual samples expand the capacity of training samples,it reduces the disturbance caused by face expression and changing posture.Then select more appropriate training samples to reconstruct the test sample,and finally using the results of the reconstructing to identify the classification of the test sample.Some face recognition experiments show that this method can realize lower face recognition error rate.In order to solve the problem of insufficient number of training samples which caused by variant illumination,this paper proposed sparse representation method based on virtual sample and Gauss kernel distance produces virtual samples.This method by adding random noise to original training samples to generate virtual samples,which can simulate the possible view if face image under the condition of uneven illumination and illumination changes.Because expanded the number of training samples,it reduces the disturbance caused by uneven illumination and illumination changes.Then it uses Gauss kernel distance to select the training samples for the test sample,and then uses the linear combination of these training samples to represent the test sample,and finally finishes the classification of the test sample.Some face recognition experiments show that the two methods this paper proposed can realize lower face recognition error rate.