Home // International Journal On Advances in Security, volume 17, numbers 1 and 2, 2024 // View article
Authors:
Dennis Siegel
Christian Kraetzer
Stefan Seidlitz
Jana Dittmann
Keywords: forensics; media forensics; DeepFake detection; machine learning.
Abstract:
DeepFake technology poses a new challenge to the validation of digital media integrity and authenticity. In contrast to `traditional' forensic sub-disciplines (for example dactyloscopy), there are no standardized process models for DeepFake detection yet that would enable its usage in court in most countries. In this work, two existing best-practice methodologies (a data-centric model and a set of image authentication procedures) are combined and extended for the application of DeepFake detection. The extension includes aspects required to expand the focus from digital images to videos and enhancements in the quality assurance for methods (here focusing on the peer review aspect). Particular emphasis is put on the different actors involved in the forensic examination process. The new methodology is applied to the example of DeepFake detection in two application scenarios, based on image and video respectively. The process itself is further separated in the initial assessment of the media followed by DeepFake detection. In total 36 features from nine existing and implemented tools are used as methods. In addition, the value types, ranges and their tendency for a DeepFake are determined for each feature. To further diversify the application field, the DeepFake detectors represent both hand-crafted and deep learning based feature spaces for Media content analysis. The whole process is then manually evaluated, highlighting potential loss, error and uncertainties within the process and individual tools. With the discussed potential extensions towards video evidence and machine learning involved, we identified additional requirements. These requirements are addressed in this paper as a proposal for an extended methodology to serve as starting point for future research and discussion in this domain.
Pages: 29 to 43
Copyright: Copyright (c) to authors, 2024. Used with permission.
Publication date: June 30, 2024
Published in: journal
ISSN: 1942-2636