A Systematic Review On Generative Adversarial Networks (GANs) For Biometrics

Abstract

Deep learning, a subset of machine learning, aims to imbue machines with human-like perception, learning, and intelligence through advanced technology. It has made significant strides in fields such as speech recognition, computer vision and NLP. GAN (Generative Adversarial Network) is an innovative domain within deep learning that has made significant growth in areas like image processing, art, music, data analysis, drug discovery, and gaming industry applications. Biometric systems, which use distinct physiological features like iris patterns, fingerprints, and facial characteristics for authentication, have increasingly integrated deep learning models. As biometric technology becomes widespread, ensuring information security in sectors like banking, education, and airports is paramount. Deep learning-based generative networks have revolutionized synthetic biometric data generation, produced high-quality artificial data while preserved the statistical characteristics of the original dataset. These synthetic biometric datasets are invaluable for testing and developing biometric systems, especially under high-demand conditions. The purpose of this paper is to present a broad overview of GANs, its loss function, highlighting the popularly used architectures and application domains of the most well-known variations. The optimal biometric application area and the efficacy of various model designs will be discussed.