Cyber Vaccine for Deepfake Immunity
Ching‐Chun Chang, Huy H. Nguyen, Junichi Yamagishi, Isao Echizen
Abstract
Deepfakes pose an evolving cybersecurity threat that calls for the development of automated countermeasures. While considerable forensic research has been devoted to the detection and localisation of deepfakes, solutions for ‘fake to real’ reversal are yet to be developed. In this study, we introduce the concept of cyber vaccination for conferring immunity to deepfakes. In other words, we aim to impart self-healing ability to the face media so that the original content is possible to be recovered after manipulated by AI-based deepfake technology. Analogous to biological vaccination which uses injected antigens to induce immunity prior to infection by an actual pathogen, cyber vaccination simulates deepfakes and performs adversarial training to build a defensive immune system. Aiming at building up attack-agnostic immunity with limited computational resources, we propose simulating various deepfakes with one single overpowering attack: face masking. The proposed immune system consists of a vaccinator for inducing immunity and a neutraliser for recovering facial content. Experimental evaluations demonstrate effective immunity to face replacement and various types of corruptions.