Sparse Combined Index-of-Max Hashing for Fingerprint Template Protection

Abstract

In this paper, a cancellable biometric scheme called Sparse Combined Index-of-Maximum (SC-IoM) hashing is proposed. SC-IoM hashing is designed on top of IoM hashing proposed recently. Unlike IoM hashing, SC-IoM hashing acquires indices of the largest and second largest user-specific random projected biometric features, thus produce two set integer-valued vectors. Two hash vectors are further nonlinearly transformed into sparse binary vectors via a probabilistic many-to-one function, which is manifested in a Lookup Table form. Finally, two sparse binary vectors are mixed intersectly and yield a compact single sparse binary vector. The security and privacy of the SC-IoM hashing are enhanced over IoM hashing yet provides better accuracy performance. Both revocability and unlinkability criteria remain satisfied. The scheme is evaluated with fingerprint vectors under FVC2002 and FVC2004 fingerprint benchmark datasets.