Research On Privacy-preserving Face Recognition Algorithms

Face recognition has been a promising artificial intelligence technique and is becoming increasingly popular in civilian and law enforcement applications that require reliable biometric authentication,such as cell phone unlocking,surveillance systems,security checks,and other places.However,the wide deployment of face recognition systems has raised serious privacy concerns regarding the abuse of users’ face images.In order to protect the privacy of the user’s faces,this paper works on privacy-preserving face recognition algorithms,and the main contributions are as follows:1.We propose a privacy-preserving face verification algorithm.In this paper,we propose a privacy-preserving face verification algorithm based on a hybrid computational approach of fully homomorphic encryption and garbled circuit.Our algorithm works under the client-servercryptographic service provider(CSP)three-party model.First,the client encrypts and uploads images using Single Instruction Multiple Data Stream technology,reducing the amount of communication and improving computational performance.Second,the server performs linear operations based on fully homomorphic encryption and runs a garbled circuit with the assistance of CSP to perform nonlinear computations.Finally,the client learns nothing beyond the verification result.This algorithm is the first privacy-preserving face verification protocol such that the server cannot learn any information about face images,intermediate results,and verification result,and the client learns nothing beyond the verification result.Experimental results show that the algorithm takes less than one second to perform a privacy-preserving face verification process.2.We propose a privacy-preserving face identification algorithm.In this paper,we propose a privacy-preserving face identification algorithm based on a hybrid computational approach of fully homomorphic encryption and garbled circuit.Our algorithm works under the client-servercryptographic service provider(CSP)three-party model.First,we introduce a novel column-wise strategy that encrypts each column of the face templates database into a single ciphertext based fully homomorphic encryption technique.This strategy significantly reduces ciphertext expansion of face templates database.Second,the server performs the face identification process over ciphertext and runs the garbled circuit protocol with the CSP,and returns the identification result to the client.This work is the first privacy-preserving face identification protocol such that neither the server nor the CSP obtains probe face and face templates,access pattern,intermediate results and identification result,while providing face identification service.Compared with the existing face identification algorithms,our proposed face identification algorithm is more secure and efficient.