Research On Privacy-preserving Technology Of Edge Computation-Based Face Verification System For Authentication

In recent years,with the continuous upgrading of computer hardware,artificial intelligence technology has made a new breakthrough in practical application.Human life has become more intelligent and automated.Our society is gradually entering a pluralistic era in which everything is interconnected.Due to the unique,distinguishable,flexible and direct friendly characteristics of biological characteristics,traditional access control and identity authentication technology based on password authentication are gradually replaced by authentication schemes based on biological characteristics identification technology in many application fields.Compared with other biometric recognition technologies,face recognition technology has become the most widely used technology in the field of identity authentication due to its non-contact,non-aggression,support for infrared and visible light,and no user cooperation.Face recognition technology based on outsourcing calculation is a typical application at present.However,cloud servers provided by cloud service providers are mostly ”semitrusted” and face data are extremely sensitive,so uploading unencrypted face data directly to cloud server to request services will bring the risk of privacy disclosure.Therefore,the research focuses on how to realize the function of face recognition under the premise of protecting the privacy of outsourced face data.Recently,although the solutions based on homomorphic encryption and garbled circuit technology have made great breakthroughs in theory,there are still problems of large computation and communication overhead in practical applications.Based on the above situation,this thesis proposes a face identity authentication system based on edge computation that can protect user privacy.The contributions of the thesis are as follows:1.This thesis for the first time introduces edge computation into traditional face authentication system based on outsourcing computation,which can not only reduce the number of interactions between users and cloud servers,improve the real-time performance and fault tolerance rate of the system,but also facilitate the implementation of privacy protection schemes.In this thesis,a convolutional neural network for face image feature extraction is trained by caffe framework,and cosine similarity is used to complete face verification.Experimental results show that the face recognition scheme implemented in this thesis achieves higher accuracy than traditional face recognition algorithms on LFW data sets.2.This thesis proposes a privacy protection scheme based on the secure nearest neighbor algorithm,which is used to protect the security of the user's feature vectors registered locally in the edge computing node and to realize the edge computing node's similarity measurement operation for encrypted face feature data.The encryption process will not consume large computing resources.The experimental results show that the scheme realizes the same recognition accuracy of face feature vectors in ciphertext state and plaintext state.3.This thesis proposes a privacy protection scheme based on secret sharing homomorphism technology.Each edge computing node in the system stores the secret shadows of the user's face feature vector registered on other edge computing nodes,so that each edge computing node can cooperate with any other t-1 edge computing nodes to complete the operation of requesting to the cloud server for the user's rights registered on other edge computing nodes without disclosing the information of user's face feature vector and identity,thus realizing the cross authentication between the user and the edge computing nodes.Finally,the experiment proves that the accuracy of face recognition realized by this scheme is exactly the same as that in plaintext.