Role of human physiology and facial biomechanics towards building robust deepfake detectors: A comprehensive survey and analysis

Abstract

AI based multimedia content generation, already having achieved hyper-realism, deeply influences human perception and trust. Since emerging around late 2017, deepfake technology has rapidly gained popularity due to its diverse applications, raising significant concerns regarding its malicious and unethical use. Although many deepfake detectors have been developed by forensic researchers in recent years, there is an urgent need for robust detectors that can overcome demographic, social, and cultural barriers in identifying deepfakes. To identify a human as a human, to distinguish a person from a synthetic entity, the literature faces compelling necessity to introduce deepfake detectors that can withstand all forms of demographic and social biases. (Multiple researches have been conducted in recent times to prove the existence of social and demographic biases in synthetic media detectors.) In this article, we examine human physiological signals as the foundation for robust deepfake detectors, and present a survey of recent developments in deepfake detection research that relies on human physiological signals and facial biomechanics. We perform in-depth analysis of the techniques to understand the contribution of human physiology in deepfake detection. Hence, we comprehend how human physiology based deepfake detectors fare by exploiting the inherent robustness of physiological signals, in contrast to other existing detectors.