Robust watermarking against face swapping

Paper Title:

Dual Defense: Adversarial, Traceable, and Invisible Robust Watermarking against Face Swapping

Published on:

25 October 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Yunming Zhang,

Dengpan Ye,

Caiyun Xie,

Long Tang,

Chuanxi Chen,

Ziyi Liu,

Jiacheng Deng

Bullets

Key Details

•

Dual Defense embeds adversarial and traceable watermarks into facial images

•

It disrupts face swapping models while preserving watermark integrity

•

The method balances invisibility, adversariality, and traceability

•

It shows excellent robustness and cross-task universality

•

Dual Defense outperforms prior defense methods in adversariality and traceability

Explore the topics in this paper

adversarial attacks

cybersecurity

facial recognition

forensics

image watermarking

AI generated summary

Robust watermarking against face swapping

This paper proposes a new method called Dual Defense that embeds invisible watermarks into facial images. The watermarks disrupt face swapping models, while still allowing extraction of the watermark for tracing image provenance. Dual Defense balances invisibility, adversarial attack success, and traceability.