A Comprehensive Survey of Generative Adversarial Networks in Biometric Applications

Abstract

Being a sector of machine learning, deep learning aims to give robots abilities much like humans in learning, perception and intelligence. In the fields of image processing, GAN (Generative Adversarial Network) has become an important branch of deep learning. Now, deep learning generative networks allow for the generation of good quality synthetic data that still keep the same statistics as the original data. Keeping information secure in various important sectors like education, banking and others is very important as biometrics are adopted more and more for authenticity. During periods of large demand, synthetic biometric datasets come in very useful for checking and developing biometric systems. The amount of high-quality data, the training process’s reliability and updates to the architecture are part of the evaluation too. Generative Adversarial Networks (GAN) is one of the most reliable frameworks for making realistic synthetic data. This study reviews several GAN architectures; the loss function they use and the common designs as well as fields of application and also looks at how they help in biometrics. Using unique characteristics from the body like fingerprints, face details and iris patterns, biometric systems now commonly use deep learning models. The research also looks at how GAN variations can be used in face, iris, fingerprint and palmprint identification. The purpose of this paper is to discuss the GAN-based methods in biometrics by systematically comparing these models and how they are used.