A Systematic Review of Deep Learning Approaches for Synthetic Media Detection in Biometric Authentication Systems

Abstract

The recent advancements in synthetic media generation technologies have brought about various challenges in the reliability of facial biometric authentication systems in security sensitive areas. This systematic review explores current studies pertinent to deep learning-based solutions for analyzing compromised facial media, with special attention paid to their adaptability for use in biometric authentication systems. A structured analysis was applied to twenty-eight peer-reviewed studies published from 2020 to 2024, with special focus placed on solutions based on convolutional neural networks, transformer-based networks, and liveness detection methods for facial anti spoofing. Results show that transformer-based solutions demonstrate outstanding detection capability and resistance against intricate manipulation patterns, while convolutional neural network-based solutions possess lower computational complexity and adaptability for real-time authentication applications in biometric systems. Yet both demonstrate shortcomings in generalization capability across varied data sets and susceptibility to ever-advancing synthetic media generation technologies. Liveness detection is recognized as a supplementary mechanism for enhancing security for these systems despite increased complexity in implementation and infrastructure requirements. This systematic review draws attention to existing knowledge gaps in current studies for the development of more secure, efficient, and adaptable deep learning-based facial biometric authentication systems sensitive to novel threats from synthetic media generations.