Dual-Rule Watermarking With Facial Features of Topological and Semantic for Deepfake Proactive Defense
IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 17495-17509
DOI: 10.1109/tcsvt.2026.3722506 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Yongjin Xian und 3 weiteren Autor:innen. Die Arbeit nennt 67 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Shandong Provincial Natural Science Foundation, Quan Cheng Laboratory.
Abstract (Originalsprache)
With the rapid development of Deepfake generation techniques, forged facial images have become increasingly realistic, posing serious risks such as privacy leakage and the dissemination of misinformation. Existing passive detection methods operate in a post-hoc manner and often exhibit limited generalization to unseen manipulations. Meanwhile, existing proactive watermarking methods commonly rely on content-independent random bits and separately designed encryption strategies, resulting in limited interpretability. To address these limitations, we propose a proactive watermarking framework for facial images that constructs a dual-rule watermark from facial topological constraints and semantic attributes. Specifically, inspired by quantum control, this paper introduces a learnable encryption module to achieve adaptive encoding and secure protection of watermark information, thereby enhancing the imperceptibility and resistance to interference during watermark embedding. On this basis, a feature-consistency-based detection strategy is further designed, which models the consistency between original features and forged image features to facilitate Deepfake detection. Furthermore, the watermark embedding and extraction processes are jointly optimized to improve the robustness of watermark information while preserving the visual quality of facial images. Experimental results demonstrate that the proposed method maintains a high watermark recovery rate under common distortions and attacks, and effectively supports Deepfake image detection and defense.
Förderung
- National Natural Science Foundation of China
- Shandong Provincial Natural Science Foundation
- Quan Cheng Laboratory
- Shandong Provincial Universities Youth Innovation Technology Program
- Jinan University “20 Measures” Innovation Team Project
Zitieren
APA
Xian, Y., Xue, S., Ma, B., & Qu, Y. (2026). Dual-Rule Watermarking With Facial Features of Topological and Semantic for Deepfake Proactive Defense. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 17495-17509. https://doi.org/10.1109/tcsvt.2026.3722506
Vancouver
Xian Y, Xue S, Ma B, Qu Y. Dual-Rule Watermarking With Facial Features of Topological and Semantic for Deepfake Proactive Defense. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):17495-17509. doi:10.1109/tcsvt.2026.3722506
BibTeX
@article{Xian2026,
author = {Xian, Yongjin and Xue, Shihui and Ma, Bin and Qu, Youyang},
title = {Dual-Rule Watermarking With Facial Features of Topological and Semantic for Deepfake Proactive Defense},
journal = {IEEE Transactions on Circuits and Systems for Video Technology},
year = {2026},
volume = {36},
number = {11},
pages = {17495--17509},
doi = {10.1109/tcsvt.2026.3722506}
}
EUVIMED