Published on:
2 November 2023
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Tianyi Wang,
Mengxiao Huang,
Harry Cheng,
Bin Ma,
Yinglong Wang
Proposes identity perceptual watermarks with chaotic encryption for deepfake defense
Develops robust encoder-decoder framework to embed watermarks into facial images
Achieves concurrent deepfake detection and source image tracing
Demonstrates strong performance across datasets and face swapping algorithms
Outperforms existing watermarking and distortion methods for deepfake defense
Detecting and tracing deepfakes using identity watermarks
This paper proposes a new method to detect deepfake face swapping manipulations and trace the original image source. It embeds imperceptible identity watermarks into facial images that contain semantic information about the face identity. A chaotic encryption system protects watermark confidentiality. The watermarks are encoded and recovered using an encoder-decoder neural network framework trained to be robust against common image manipulations and face swapping attacks. Experiments show this method has state-of-the-art performance in recovering watermarks and detecting fakes, even on unseen datasets and face swapping algorithms.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper