Privacy preservation for soft biometrics based multimodal recognition system

Abstract

There has been a recent proliferation in the utilization of biometric systems as an access granting mechanism for various applications. Researches have demonstrated that the overall performance of these recognition systems significantly improves upon the inclusion of “soft” biometric traits in addition to the conventional primary biometric modalities. Normal biometric systems can be subjected to a wide variety of malicious attacks and privacy breaches, which ultimately lead to severe jeopardy for the enrolled users. These security infringements are generally concentrated on a central database wherein the user's biometric data get stored. The privacy risks associated with soft biometric aided designs comprehensively increase due to the availability of more personalized amount of information in these databases. In our work we have addressed this problem and proposed a privacy preserving framework for soft biometric based systems by utilizing the popular notion of differential privacy. We have also analyzed our scheme on real life data and found that the privacy preservation technique does not affect the utility (performance) of the basic biometric system. Thus our proposed framework provides privacy for biometric system users without compromising its overall performance.