Overview of Technologies for Detecting Modified Content of the DeepFake Class

Abstract

The article provides an overview of publications on the main technologies for detecting modified content of the DeepFake class, including a list of publicly available datasets and the results of testing solutions available in the public domain. The article also presents the results of independent testing of DeepFake detection systems, obtained during the DFDC ‒ 2019 (DeepFakeDetectionContest) and gives brief reviews of similar competitions held in 2020-2021 in China. The article also describes a new approach to detecting photo/video materials created using DeepFake technologies. The authors' analysis of the solutions made it possible to create a new way of detecting fake content, which was patented by the authors in the Federal Service for Intellectual Property for Technology. The DeepFake detection mechanism proposed by the authors has a number of significant advantages over the solutions described in the review, which allows us to count on more efficient detection of attempts to bypass biometric systems.

Keywords

• deepfake
• biometric presentation attacks
• facial biometrics