Research On Face Recognition Of Image Set Based On Complex Network

In the field of traditional face recognition,a single static facial image is usually used for analysis and classification.Most methods have achieved good recognition performance in controllable scenes.However,these algorithms based on a single image perform poorly in many practical applications,since the obtained facial image under unconstrained conditions is easily affected by the variations in illumination condition,facial expression,pose and other factors.With the recent advances in digital imaging and communication technologies,multiple images(i.e.image set)of a person are easily collected.Therefore,the testing image sets can be compared with the training image sets in the procedure of face recognition to improve the accuracy of classification.Researchers formulate this type of face recognition(i.e.face recognition through multiple images)problem as image set based face recognition(ISFR).Generally,ISFR uses image sets for analysis and classification.Image set can more comprehensively depict a wide range of variations in expression,pose and illumination condition etc.,and ISFR can eliminate the influence of these interference factors.Therefore,in contrast to traditional face recognition methods based on a single facial image,ISFR methods can achieve more robust classification.Present face recognition approaches based on image set data lack effective feature description for image set.Besides,it is not sufficient in studying the relationships among image set data manifolds.To address these problems,a novel method based on complex network and sparse representation is proposed for the task of ISFR in this paper.This method employs the acquired facial image set data under uncontrolled conditions to construct complex network(each facial image sample in the facial image set represents a node in the network and the mutual relationship between the samples represents the edges in the network),and uses complex network method to divide the date manifold of facial image set into multiple sub-manifolds with their own characteristics;And then,based on the sub-manifolds of facial image set data,build sparse-embedding graph model and optimal projection objective function to dig theintrinsic relationships among sub-manifolds deeply and preserve the low dimensional manifold structure of face images.Finally,sub-manifold based face recognition of image set is implemented in low dimensional space.In order to verify the effectiveness of the proposed method,in this paper,we designed and conducted experiments on three well-known face image set database(i.e.Honda/UCSD,CMU Mo Bo and You Tube Celebrities).The experimental results show that the proposed method achieves better recognition performance than the state-of-the-art approaches such as sparse approximated nearest points(SANP),regularized nearest points(RNP)and joint regularized nearest points(JRNP)etc.,which demonstrates its superiority.Therefore it is suitable for ISFR.