A Locality Sensitive Hashing Based Approach for Generating Cancelable Fingerprints Templates

Abstract

Cancelable biometric schemes have emerged as a promising approach for providing robust security guarantees to extracted biometric features. In this paper, we develop a working framework for generating secure templates from raw fingerprint images utilizing the notion of Locality Sampled Codes (LSC). For achieving such objectives, we initially represent the features of a fingerprint image in a binarized form, and subsequently generate the final cancelable template by sampling random bit locations from it. Since the LSC technique is functionally established on the principle of Locality Sensitive Hashing (LSH), the induced transformations do not degrade the performance of the overall biometric model. We have performed a thorough theoretical analysis coupled with comprehensive empirical justifications for investigating the fulfillment of properties like non-invertibility, revocability, and unlinkability. We have also analyzed the performance of the model over the FVC2002-DB1, FVC2002-DB3, FVC2004-DB1, and FVC2004-DB3 fingerprint databases, for which we have obtained comparatively low EERs of 0.19%, 1.44%, 1.28% and 2.72% respectively in the stolen-token scenario.