FractalForensics was jointly proposed by a research team from the National University of Singapore and Shandong University in April 2025, and the relevant research results were published in a paper. FractalForensics: Proactive Deepfake Detection and Localization via Fractal WatermarksIt has been selected for the ACM Multimedia 2025 Oral.

FractalForensics is a novel approach for active deepfake detection and localization. Leveraging the properties of fractal shapes, researchers first designed a parameter-driven pipeline for watermark generation, allowing user-selectable parameters to determine variations in the fractal shape and watermark encryption rules via a chaotic cryptographic system. An innovative position-aware watermark embedding and recovery element-to-patch strategy is introduced when placing 4-bit watermark matrix elements along the channel dimension, implicitly aiding in the interpretable localization of malicious deepfakes. FractalForensics achieves superior robustness and vulnerability in terms of watermark recovery rate, resulting in promising performance for active deepfake detection and localization, surpassing state-of-the-art passive and semi-fragile methods.