Providing robust security measures to Bloom filter based biometric template protection schemes

Abstract

Template protection is an essential requirement of biometric recognition systems. These special methods are designed to provide the necessary security and privacy privileges to the registered users of the biometric system. Similar to other security domains, these schemes require to follow certain properties or characteristics like unlinkability, irreversibility and low information leakage from the stored data. Ideally, these schemes must not result in any degradation of the biometric recognition performance as a whole. Designing a system which balances all these factors efficiently has been a major challenge. Contemporary frameworks generally relax one or more of these parameters in order to achieve their application specific goals. However recently, a modified Bloom filter based scheme was proposed which apparently provided competent security measures in addition to high recognition accuracy rates. However subsequent studies refuted the security guarantees of the scheme by providing adversarial attacks with practical bounds. In our work, we address this issue and propose a modified Bloom filter based framework which provides all the desirable security measures of a biometric template protection scheme. Not only have we theoretically proved all the security notions of our model, but also experimentally verified our claims. Thus our proposed design provides satisfactory security guarantees in addition to the implicit advantages of using Bloom filters like speed and efficiency. Finally, there is no reduction in the performance rates of the underlying system since we perform the matching of the biometric traits in their original forms (in contrast to some transformed space).