Process-Driven Modelling of Media Forensic Investigations-Considerations on the Example of DeepFake Detection

Area of Science:

• Digital Forensics
• Media Forensics
• Artificial Intelligence

Background:

• Academic media forensics research focuses on manipulation detection, but lab performance rarely translates to courtroom readiness.
• A significant gap exists between academic research and the practical requirements of forensic practitioners.

Purpose of the Study:

• To bridge the gap between media forensics researchers and practitioners.
• To facilitate mutual understanding and propose steps toward closing this gap.
• To model media forensic investigation pipelines for practical application.

Main Methods:

• Derived a concept for modeling media forensic investigation pipelines based on established guidelines.
• Illustrated pipeline modeling using a fusion-based approach for DeepFake video detection.
• Tested five exemplary detectors and two fusion operators within the pipeline model.

Main Results:

• The study models a media forensic investigation pipeline, demonstrating its applicability for DeepFake detection.
• Fusion of multiple detectors within a structured pipeline shows potential for enhanced forensic analysis.
• The approach facilitates the integration of AI-based methods into practical forensic investigations.

Conclusions:

• Modeling investigation pipelines is crucial for translating academic media forensics research into courtroom-ready tools.
• A structured, fusion-based approach can improve the reliability and applicability of DeepFake detection methods.
• This work provides a framework for developing and validating AI-driven forensic investigation methods.