Big Brother is Watching: Proactive Deepfake Detection via Learnable Hidden Face

Authors: Hongbo Li, Shangchao Yang, Ruiyang Xia, Lin Yuan, Xinbo Gao

Published: 2025-04-15 15:50:54+00:00

AI Summary

This paper introduces a proactive deepfake detection framework that embeds a learnable "hidden face" into a host image using a semi-fragile invertible steganography network. This hidden face acts as an imperceptible indicator, monitoring for malicious forgery when restored by the inverse process. The secret template is optimized during training to resemble a neutral facial appearance, enabling robust detection of tampering while maintaining imperceptibility against benign image processing.

Abstract

As deepfake technologies continue to advance, passive detection methods struggle to generalize with various forgery manipulations and datasets. Proactive defense techniques have been actively studied with the primary aim of preventing deepfake operation effectively working. In this paper, we aim to bridge the gap between passive detection and proactive defense, and seek to solve the detection problem utilizing a proactive methodology. Inspired by several watermarking-based forensic methods, we explore a novel detection framework based on the concept of ``hiding a learnable face within a face''. Specifically, relying on a semi-fragile invertible steganography network, a secret template image is embedded into a host image imperceptibly, acting as an indicator monitoring for any malicious image forgery when being restored by the inverse steganography process. Instead of being manually specified, the secret template is optimized during training to resemble a neutral facial appearance, just like a ``big brother'' hidden in the image to be protected. By incorporating a self-blending mechanism and robustness learning strategy with a simulative transmission channel, a robust detector is built to accurately distinguish if the steganographic image is maliciously tampered or benignly processed. Finally, extensive experiments conducted on multiple datasets demonstrate the superiority of the proposed approach over competing passive and proactive detection methods.

Key findings

The proposed approach demonstrates superior deepfake detection performance across multiple datasets and diverse deepfake techniques, outperforming both passive and other proactive methods. It exhibits strong robustness against various benign image manipulations while maintaining high detection accuracy. Additionally, the method achieves significantly better steganographic image quality compared to competing proactive approaches, ensuring imperceptible embedding of the hidden face.

Approach

The method utilizes a semi-fragile invertible steganography network to embed a learned, neutral-looking face template into a host image. This steganographic image is then released through a simulated transmission channel where it undergoes benign or malicious manipulations. A deepfake detector, based on a patch discriminator, identifies forgery by analyzing the differences between the restored secret template and the original, leveraging the network's sensitivity to malicious alterations and robustness to benign ones.

Datasets

FFHQ, VGGFace2, CelebA-HQ

Model(s)

Invertible Neural Network (INN) with DWT, IWT, and Affine Coupling Blocks (ACBs) for steganography; Patch discriminator for deepfake classification; Self-blending mechanism (SBI) for malicious forgery simulation.

Author countries

UNKNOWN

← Previous