A Framework for Cloud Data Security Using Fingerprint Biometrics and Homomorphic Encryption Technology
Abstract
In this paper, we evaluate a new privacy-preserving scheme for cloud authentication combining a biometric verification using fingerprint, a novel challenge-response procedure, and additive homomorphic encryption. The scalability of cloud computing allows us to store and process huge amounts of data without need to maintain locally large computational resources; but at the same time, this approach brings new security problems associated with authentication procedures involving sensitive data. Since unlike passwords, fingerprints are not disposable after compromise, they should be additionally secured. The proposed cloud authentication scheme was tested via an experiment using SOCOFing fingerprint dataset. Preprocessing steps include image check, normalization, enhancement, and conversion to minutiae-based numeric templates. Encrypted versions of the registered templates are created using Paillier cryptosystem before being uploaded to the cloud. In the authentication phase, the same procedure is performed for the query fingerprint, the new challenge is checked and supported cloud computation is carried out for the encrypted data. Finally, the matching decision is decrypted under control of the user and compared to the predefined biometric threshold. A first-in-first-out queue approach is used to analyze the behavior of authentication requests under the cloud workload conditions along with possible replay and spoofing attacks. According to our experimental results, the accuracy of fingerprint authentication equals 98.02% (false acceptance rate = 0.0033, false rejection rate = 0.0002). The average delay caused by the challenge-response phase is equal to approximately 47.8 ms. We would also like to mention the expected cost of such implementation in terms of privacy. Namely, the computational complexity of encryption is higher than that of decryption and the size of encrypted templates exceeds the size of plaintext templates.