Similarity Learning for multimodal Biometrics

With the rapid use of technology and advancements in various domains of digital services, everyone needs secure and accurate identity verification systems. Traditional unimodal biometric systems generally suffer with the issue of inaccurate data, spoofing attacks and the irregularities introduced because of unconstrained environment. One of the solutions of these issues is to take one step ahead and move to multimodal biometric authentication system. These systems rely on multiple biometric traits instead of one at the time of validating the person’s identity. Earlier researchers used to rely on handcrafted features extracted from the images for matching, later with the advancement of technology deep learning concepts transformed this domain completely. Now advanced deep learning architectures such as Siamese network and transformer models are providing a powerful framework for verification. This paper presents a deep discussion on the role of Siamese Networks in multimodal biometric authentication. The facts highlighted in the paper are based on a detailed examination of recent developments in similar domain.