Research And Implementation Of Digital Identity Technology Based On Face Template Protection

The rapid development of the Internet has made biometric technology widely used in society,but at the same time,some people have begun to notice that there are many security and privacy issues in their biological template information.Reseachers have proposed many protection algorithms for biological template information,but they generally need to sacrifice identifiability in exchange for high security.The tremendous breakthrough of deep learning technology has brought rich possibilities to biometrics,but it also led that illegal researchers use deep learning models to create various non-existent fake face images.This illegal behavior seriously pollutes a large number of face datasets,which has a serious impact on the accuracy of biometric recognition.It needs to ensure that the input data is a real face image to have a better recognition accuracy for biometrics.Therefore,it is an important research content to combine the advantages of deep learning technology with face template protection and counterfeit face detection.This article focuses on face template protection and fake face detection.The main research work includes the following:(1)In order to improve the recognizability while ensuring high security,this paper proposes a face template generation network Binary Face based on deep convolutional neural network and random orthogonal mapping,and proposes a two-stage Face template protection scheme.It will ouput a cancelable binary template after Binary Face dealing with the user's registered face image,and then enter the key generated in the random orthogonal mapping into the fuzzy commitment scheme to generate an encrypted face template and store it in the database.In the verification phase,the key is recovered through the same process and compared with the original key to obtain the final matching score.In terms of feature conversion,in order to facilitate mobile deployment,Binary Face has been light-weighted designed.At the same time,the random mapping in the traditional three-stage method is changed to a random orthogonal mapping and merged into the Binary Face network.While implementing the end-to-end training of the model,it greatly improves the operating efficiency;in the aspect of biological encryption,the traditional face template protection scheme does not consider the impact of the quantization loss of the binary template on the recognizability when implementing encryption.This paper designs Binary Loss to minimize the Quantization Loss caused by converting real-valued templates.The two-stage face template protection scheme was evaluated in three datasets including CMU-PIE,FEI,Color FERET.Compared with the previous work,it has about 6.5% improvement in GAR,while reducing EER by about 4 times.(2)In order to solve the problem of authenticity of the user's face dataset,this paper designs a light-weighted forged face detection network based on Xception Net and deep separable bottleneck structure.In order to ensure the diversity of the dataset,on the one hand,four classic fake face generation models are used to create the database,on the other hand,three different video formats(original video,high-quality video,low-quality video)are set to improve the generalization ability of the Fake face detection model.This article compares the improved Xception Net model with the remaining five classic models,and all models are retrained on the five datasets.The improved Xception Net model is more efficient than other models while improving accuracy.