Palmprint template protection scheme based on randomized cuckoo hashing and MinHash

Abstract

To provide the necessary security and privacy privileges for registered users of biometric systems, a novel template protection method for generating cancelable palmprint features based on randomized cuckoo hashing and minHash is proposed in this paper. Firstly, the orthogonal features of palmprint are extracted using the anisotropic filter. Then, a novel randomized cuckoo hashing is applied on the binary palmprint feature as a means of first layer security. Randomized cuckoo hashing composes of two hash tables; before hashing, each column of original biometric template is filtered with a random matrix to improve discriminative ability. For the randomized cuckoo hashing with the same positions, we use different Gray coding methods instead, which improves irreversibility. Furthermore, to improve the efficiency and resist unlinkability attacks, another layer of privacy protection, MinHash is adopted. Experimental results on PolyU database show that our method can nearly preserve the original verification performance. The irreversibility, unlinkability and revocability properties are examined experimentally. Also, the privacy of palmprint protection scheme is analyzed under the Brute-force Attack, False Accept Attack and Birthday attack.