FaceGuard: Proactive Deepfake Detection

View on arXiv ← Back to list

Authors: Yuankun Yang, Chenyue Liang, Hongyu He, Xiaoyu Cao, Neil Zhenqiang Gong

Published: 2021-09-13 02:36:25+00:00

AI Summary

FaceGuard is a proactive deepfake detection framework that embeds robust yet fragile watermarks into real face images. It detects deepfakes by comparing extracted watermarks to ground truth, flagging inconsistencies as fake.

Abstract

Existing deepfake-detection methods focus on passive detection, i.e., they detect fake face images via exploiting the artifacts produced during deepfake manipulation. A key limitation of passive detection is that it cannot detect fake faces that are generated by new deepfake generation methods. In this work, we propose FaceGuard, a proactive deepfake-detection framework. FaceGuard embeds a watermark into a real face image before it is published on social media. Given a face image that claims to be an individual (e.g., Nicolas Cage), FaceGuard extracts a watermark from it and predicts the face image to be fake if the extracted watermark does not match well with the individual's ground truth one. A key component of FaceGuard is a new deep-learning-based watermarking method, which is 1) robust to normal image post-processing such as JPEG compression, Gaussian blurring, cropping, and resizing, but 2) fragile to deepfake manipulation. Our evaluation on multiple datasets shows that FaceGuard can detect deepfakes accurately and outperforms existing methods.

Key findings

FaceGuard outperforms existing passive detection methods, achieving high accuracy across multiple datasets and deepfake generation techniques. It remains effective even against adaptive deepfakes that attempt to counter the watermarking approach. The embedded watermarks do not significantly impact image perceptual quality.

Approach

FaceGuard embeds watermarks into real face images using a deep learning-based encoder-decoder model. It then extracts these watermarks from potentially manipulated images and compares them to the originals; a mismatch indicates a deepfake.

Datasets

FaceForensics++, Facebook Deepfake Detection Challenge (DFDC), Trump-Cage dataset

Model(s)

Convolutional Neural Networks (CNNs) for encoder, decoder, and discriminator. Xception network used for comparison passive detector.

Author countries

China, USA

← Previous