Research On New Face Recognition Algorithms Based On PCANet And Sparse Representation

Face recognition technology,as a popular research in the field of computer vision,has a long history and a wide range of applications in various industries.In recent years,sparse representation has gained increasing attention in computer vision.Methods based on sparse representation,such as sparse representation classification,have produced good results in face recognition.In order to further improve the classification accuracy of face recognition,four improved face recognition algorithms are proposed in this paper.The contributions of this paper can be summarized as follows:First,a face recognition algorithm using PCANet to improve dictionary construction in sparse representation is proposed in this paper.Principal component analysis network(PCANet)is a newly proposed deep learning algorithm.Compared with other convolutional neural networks,it has the advantage of simple network architecture and easy training.In this algorithm,PCANet is used to extract new features from face images,and then the extracted new features are used to construct dictionary for sparse representation.And then,it is found that in the traditional face recognition algorithm,the Euclidean distance is generally used as the measure of face similarity in the classification stage,in this paper,the face image correlation used for classification is proposed.The definition and calculation process of face image correlation were introduced after that,then face recognition algorithm based on PCANet and correlation analysis is proposed.Second,the existing face recognition algorithms based on sparse representation usually use the reconstruction error from one class for classification,but ignore the other classes.It would be better if the sample(y)to be classified is more like the samples in the assigned class(if it is currently to determine whether y belongs to the i-th class,the i-th class is the assigned class),and at the same time it is less like the samples in the other classes(all classes except the i-th class).So the dual reconstruction errors are introduced in this paper,and then face recognition algorithm using dual reconstruction errors to improve sparse representation classification and face recognition algorithm based on PCANet and dual reconstruction errors are proposed.In order to verify the effectiveness of the four algorithms proposed in this paper,the experiments on five public face data sets were conducted.The proposed algorithms are compared not only with the current popular face recognition algorithms,but also with themselves.Experimental results show that the proposed algorithms can achieve the similar or even better classification results on most data sets.Among the algorithms proposed in this paper,the face recognition algorithm based on PCANet and dual reconstruction errors has the best performance.