WaveGuard: Robust Deepfake Detection and Source Tracing via Dual-Tree Complex Wavelet and Graph Neural Networks

View on arXiv ← Back to list

Authors: Ziyuan He, Zhiqing Guo, Liejun Wang, Gaobo Yang, Yunfeng Diao, Dan Ma

Published: 2025-05-13 14:31:42+00:00

AI Summary

WaveGuard is a proactive watermarking framework for deepfake detection and source tracing that embeds watermarks into high-frequency sub-bands using the Dual-Tree Complex Wavelet Transform (DT-CWT) and a Structural Consistency Graph Neural Network (SC-GNN) to maintain visual quality. It outperforms state-of-the-art methods in robustness and visual quality.

Abstract

Deepfake technology poses increasing risks such as privacy invasion and identity theft. To address these threats, we propose WaveGuard, a proactive watermarking framework that enhances robustness and imperceptibility via frequency-domain embedding and graph-based structural consistency. Specifically, we embed watermarks into high-frequency sub-bands using Dual-Tree Complex Wavelet Transform (DT-CWT) and employ a Structural Consistency Graph Neural Network (SC-GNN) to preserve visual quality. We also design an attention module to refine embedding precision. Experimental results on face swap and reenactment tasks demonstrate that WaveGuard outperforms state-of-the-art methods in both robustness and visual quality. Code is available at https://github.com/vpsg-research/WaveGuard.

Key findings

WaveGuard outperforms state-of-the-art methods in both robustness and visual quality on face swap and reenactment tasks. The method shows low Bit Error Rate (BER) for watermark recovery under common distortions and a high BER under malicious distortions (deepfakes), enabling effective deepfake detection. It also demonstrates good generalization to out-of-distribution data.

Approach

WaveGuard embeds watermarks in high-frequency sub-bands of images using DT-CWT for robustness. A SC-GNN maintains visual quality by ensuring consistency between original and watermarked images. Separate decoders are used for robust source tracing and semi-robust deepfake detection.

Datasets

CelebA-HQ, LFW

Model(s)

Dual-Tree Complex Wavelet Transform (DT-CWT), Structural Consistency Graph Neural Network (SC-GNN), DenseNet-like encoder and decoder architecture with attention modules.

Author countries

China

← Previous