Trustworthy biometric hashing method

In this paper, we propose a novel biometric hashing method. We employ a password-generated random projection matrix applied to the face images directly instead of applying to the features extracted from face images and improve the methods in the literature. We aim to preserve privacy while achieving desirable accuracy in a biometric verification system. We do the verification in the hash domain and ensure irreversibility. In addition, we can get a new hash value by only changing the password which ensures cancelable biometrics property. We achieve zero equal error rate (EER) on Carnegie Mellon University face database. Furthermore, we achieve an EER of 0.0061, even if the attackers compromise the password and the random number generator. Besides, we test robustness of the proposed system against possible degradations due to sensor and environment inperfections. The norm of error is below optimum threshold obtained at EER for all distortions.