Cancelable Biometrics for Template Protection

In the 21st century, deployment of large biometric systems worldwide like Aadhaar (India), eKTP (Indonesia), and MyKad (Malaysia) surges the immediate need to secure biometric systems due to their vulnerabilities to several security threats. Unlike conventional password-based systems, biometric templates, if compromised, cannot be revoked. In such situations, biometric cryptosystems and cancelable biometrics (CB) come as a rescue measure. Specifically, in the case of CB; biometric template is encoded with a distortion scheme or with a non-invertible transformation that varies for each underlying application. The concept of cancelability in biometric authentication was developed to address three non-functional requirements, i.e., revocability, non-invertibility, and unlinkability simultaneously along with functional requirement, i.e., performance. In the past decade, most of the work has been carried out to secure biometric templates, particularly for iris and fingerprints, since these are the two commonly used biometric traits for identification purposes. Apart from that, few work have been carried out on studying cancelability for multimodal biometrics and finger-veins. The fundamental issue with most of these techniques is that they are tested on a small dataset, which is mainly collected in a controlled environment. Thus, the generalisability of these techniques is questionable in the case of large-scale databases, which is indeed needed. Moreover, in the present-day scenario, attackers are getting smarter day by day, and thus in the view of new-attacks, security of already proposed techniques is quite vulnerable. An in-depth security analysis of the proposed methods is a must. As we know, there is an inexorable trade-off between security and performance; thus, one should be very careful while deciding it depending upon the application in hand. Another point of concern is that most of the techniques are applicable only on one biometric trait and fail on others except a few like BioHashing and Bloom filter which are applied simultaneously on the face, iris, and fingerprints. Since nowadays multimodal biometrics are often used, it is desirable that the same template protection technique is equally applicable with good performance to all biometric traits. Here, in this chapter, our focus will be towards illustrating different template protection techniques while keeping in mind all the above-raised issues. Moreover, the past ten years has seen a lot of research work based on deep learning techniques. Thus, our primary focus in this chapter will be to explore how the latest AI techniques, particularly deep learning-based, can shape the future of template protection. We will discuss in detail the limitations and advantages of using deep learning-based methods in the template protection domain. In addition to this, we will also suggest some future directives in the domain of template protection using cutting-edge deep learning techniques.