Robust Biometrics on Smartphones – Using Quality Assessment, Presentation Attack Detection, and Biometric Fusion

Abstract

With the technological advancements in mobile technology, there is a massive adoption of biometrics as a security measure in today1s smartphones. Smartphones are used in all day to day activities such as online banking, accessing official and personal emails, social networking and also to store personal data. Although smartphones provide high user convenience, there is an inherent security threat as losing such a device could lead to a loss of such sensitive data. This could cause disastrous effects on the smartphone user. In order to reduce the privacy and security threats, basic solutions are provided with every smartphone. However such solutions could cause user inconvenience sometimes, for example, it is hard to remember complex lock patterns, longer pin codes; also such patterns and pins could be easily hacked. Thus, an inherent need of added security measure is there and which could be conveniently fulfilled by biometrics on smartphones. As a result of which, recently, most of the smartphones are manufactured with inbuilt fingerprint sensor, or state-of-the-art face or iris recognition system. Today, we can say that for any smartphone, a biometric system is one of an essential component just like the front and rear cameras. However, the inclusion of such a biometric system comes with a cost such as the performance of a biometric system depends on several factors such as the input sample quality, systematic and random errors. Moreover, biometric systems are highly vulnerable to direct and indirect attacks. The direct attacks aka presentation attacks are carried out at the biometric sensor level by presenting a fake biometric sample. If a biometric system does not have an attack detection module also known as presentation attack detection module, it is trivial to spoof any biometric system. Thus, the primary objectives of this thesis are to address the challenges of smartphone biometrics. The unconstrained nature of biometric samples captured in a smartphone environment could cause challenging input samples for the recognition system and results in a lower comparison score. Therefore, it is essential to assess the precise quality if the input samples. In this work, we present and compare several quality assessment algorithms to formulate a unified face recognition system. This thesis proposes two presentation attack detection techniques for smartphone-based face recognition systems and one for fingerphoto recognition systems. The thesis also extends the applications of some concepts from Subjective Logic to fuse the comparison scores from face and fingerprint recognition systems. Additionally, this thesis proposes a multi-biometric and multi-algorithmic fusion scheme to mitigate the effects of body weight variations for face recognition systems. Although the proposed framework does not use smartphone biometric data, the method could be easily adapted for the smartphone-based face recognition. The validity of proposed frameworks for consistent performance is demonstrated through extensive experimentation on publicly available and newly created databases. We have also presented a new smartphone based multimodal biometric database as well as a presentation attack database in this work. Conclusively, the thesis proposes robust Biometric Quality Assessment (BQA), Presentation Attack Detection (PAD) and Biometric Fusion techniques to address the issue of sample quality assessment, presentation attacks, and multi-modal biometric fusion. A detailed experimental analysis and comprehensive studies have been executed to evaluate the proposed methods under the scope of this thesis work. The presented methods will help the researchers and users of smartphone biometrics to improve the robustness of the systems.