Research On Privacy-preserving Problems In Distributed Face Recognition System

Biometric feature recognition has become a common means of the individual identity gradually because of its characteristics such as data difficult to lost, stolen and so on.Face recognition has become a common means of biometric recognition because of its non invasive, direct and friendless, nature and convenience. The rapid development of information and communication technology such as Internet, wireless network makes the cooperative computing becoming more and more important, the centralized face recognition system can’t meet the demands of the distributed network environment; in this trend distributed face recognition system arises at the historic moment. In order to complete recognition in distributed environment cooperatively, the two sides of communication need to interact information. Because of the openness of the network environment and the sensitivity of the biometric information, its existing security threats make the privacy-preserving issue research in DFRS becoming more and more important. The privacy-preserving issue in distributed face recognition system mainly includes two aspects:(1)the private data between the client and the server can’t leak to other people or each other during recognition;(2)the server terminal face feature template protection issue. The privacy-preserving issue in DFRS is mainly including two aspects:one is the compatibility problem of cryptographic protocol and DFRS; the second is the problem of low recognition rate and low efficiency of DFRS caused by CP constraint.In view of those, the main works of this dissertation focus on privacy-preserving issue in DFRS, details as follows:(1) On the basis of detailed analysis of the Client/Server (C/S) model and the pipe-filter architecture, the architecture of DFRS was designed.(2) Based on the fully homomorphic encryption scheme over Integer,FHE-compatible privacy-preserving face recognition scheme (FHE*DFRS) was designed, protocol data units and sequence diagram in the remote pipeline were designed. Fully homomorphic encryption algorithm over integer was improved, and combining it with classifier algorithm of Euclidean distance metric to solve the compatibility problem; Gabor filter and principal component analysis were combined to extract feature to increase the recognition rate; the Euclidean distance computation was improved into the dot product computation to reduce rounds complexity.(3) Based on the fuzzy vault algorithm, a face feature template protection scheme (FV-PPFR) was proposed, which introduced Gabor filter, Principal component analysis, Binary mapping, Gaussian interference point and the Nearest searching etc. The scheme could protect the security of face feature while meeting the demands of recognition rate and efficiency.(4) The FHE*DFRS scheme and the FV-PPFR scheme were implented based on Java and OpenCV, prototype implementation and security analysis showed that the two schemes could protect privacy with high efficiency and high recognition rate.The designed two privacy-preserving schemes in the dissertation, while protecting privacy information in DFRS effectively, met the demands of high recognition rate and real-time in DFRS,and had some application value.