Privacy-preserving biometric verification with outsourced correlation filter computation

Abstract

In traditional biometric verification systems, personal computer stores biometric database and performs verification process. Because of limited storage, capacity, and computational power, both cloud computing and data centers provide these facilities for users and enterprises. However, shifting from user-owned and user-operated system to public and untrusted access services have raised concerns over data security, either in storage or computation phases. In this work, we propose a framework for fully privacy-preserving biometric verification with outsourcing computational tasks to the commerical public cloudsl. Firstly, privacy of the data used for biometric verification is preserved by encrypting training images. Secondly, for protecting the privacy of the biometric verification model, all correlation filter computation and verification stage are performed over encrypted biometric images in server side. Finally, privacy of the biometric verification result is preserved by sending it to the client for further investigation. Our solution provides anonymous access, unlinkability, and the confidentiality of transmitted data. I will be shown that our scheme is secure in the semi-honest server and has it reaches accuracy of 93.7% on facial dataset and 92% on fingerprint dataset.