The Face Validation System Based On Deep Learning

As the development of science,technology,and society,identity authentication becomes more and more important in production,living,social and company management.The convenience,reliability and efficiency of identity authentication attracts more and more people's attention.The governments,companies of all countries in the world demand urgently for convenient,reliable and efficient identity authentication technologies,and such technologies become hot research topic of many researchers in the world.To improve the performance of facial authentication system is very important for generalization of the application of the facial authentication system.The accuracy and robustness of the facial features and classifiers are key in real applications.Deep learning is a newly developed technology in artificial intelligence,which can learn and extract robust and discriminative features from unlabeled data automatically.The extracted features can represent human face in a form that suitable for recognition and authentication,which can highly improve the accuracy and time efficiency of face authentication.In this paper,after studying the traditional deep learning networks such as convolution neural network(CNN)and the principal component analysis network(PCA-Net),we propose a novel network structure for deep learning,in which we add some feature pooling and subsampling layers in the basis of PCA-Net,which improves the generalization ability on the one hand,and reduces the dimension of the output feature vector remarkably and consequently improves the times efficiency on the other hand.The main contributions of this paper are listed in the following:(1)We proposed a new deep learning network structure,which combines some features of CNN and PCA-Net.We add some feature pooling and subsampling layers in the basis of PCA-Net,and draw the advantages of the two network structures.The proposed algorithm avoid the demand of a huge number of labeled data in network training on the one hand,and largely reduces the dimension of the output structure on the other hand,which consequently highly improves the speed of recognition and authentication.Experimental results show that the propose algorithm cost fewer times in network training and the output feature vectors of the proposed algorithm outperform the output feature vectors of CNN and PCA-Net in classifying ability.(2)After extracting the facial feature vectors,we propose a algorithm of face verification based on support vector machine.The proposed algorithm highly improves the accuracy of recognition and authentication comparing with the traditional algorithms.(3)Finally,we develop a face authentication system based on the proposed deep learning feature extraction and SVM authentication algorithms.In this system,we integrate face detection,face image enhancement and face verification.