ABSTRACT
Deep learning has emerged as most effective approach for classification tasks in many biometric applications with high accuracy in classification. Among various biometric characteristics, fingervein has been employed in many applications including highly secure access control used in financial services. The difficulty in attacking fingervein-based systems is perceived to relate to the fact that vein patterns are present in the subdermal layer. On the contrary, the threat of attacking fingervein sensors was recently demonstrated with simple artefacts in printed form and electronic displays. In the wake of such threats, texture, and motion-based approaches were explored to formulate robust presentation attacks against fingervein systems. In this chapter, we systematically evaluate the presentation attack detection techniques on a relatively large-scale database of 300 unique fingervein videos. We demonstrate the applicability of texture-based approaches in the first part and then present a new approach of presentation attack detection. The newly proposed approach leverages the recent success of deep learning paradigm, and thereby, we employ transferable features obtained from the deep convolutional neural network (CNN). With the set of extensive experiments on the large dataset, we show the applicability of the proposed approach to achieve a classification error rate of 0% in detecting both the bona-fide presentation and artefact presentation of printed and electronic attacks.
