An alignment-free cancelable fingerprint template for bio-cryptosystems

Abstract

Biometric based authentication systems have been widely deployed to mobile devices. Security and privacy issues arise especially when the biometric templates stored in centralized database or personal portable devices are obtained by adversaries. To make matters worse, unlike password, biometric traits are irreplaceable once compromised. Despite the fact that a number of methods have been reported to protect biometric templates, there are rare solutions that satisfy all the criteria of template protection simultaneously (i.e. non-linkability, cancelability, non-invertibility and performance). In this paper, we propose an alignment-free cancelable template scheme to protect fingerprint minutiae. The proposed template scheme is the extended version of the polar grid based 3-tuple quantization with condensed feature length for lower computation cost. To enhance the non-invertibility, bit-toggling strategy is proposed to inject noise into the proposed fingerprint template. Furthermore, cancelability is also achieved by incorporating discrete Fourier transform and random projection on the proposed template. Due to the use of discrete Fourier transform, the resultant template will be converted into complex form which increases the difficulty of attacker in breaking its security. The proposed template scheme is evaluated on FVC2002 and FVC2004 databases (DB1, DB2 and DB3) and the experimental results demonstrate a comparable accuracy over the existing methods. The observation from the security and privacy analysis suggests that the proposed template is robust against the major attacks, e.g. template inversion attack, attack via record multiplicity, etc. Lastly, we demonstrate that the proposed template can be applied in the bio-cryptosystems.