Disruptive Attacks on Face Swapping via Low-Frequency Perceptual Perturbations

View on arXiv ← Back to list

Authors: Mengxiao Huang, Minglei Shu, Shuwang Zhou, Zhaoyang Liu

Published: 2025-08-28 09:34:53+00:00

AI Summary

This paper proposes an active defense against deepfake face swapping by introducing low-frequency perceptual perturbations to disrupt the generative process. The method combines frequency and spatial domain features to generate imperceptible artifacts, reducing the effectiveness and naturalness of deepfakes while preserving visual quality.

Abstract

Deepfake technology, driven by Generative Adversarial Networks (GANs), poses significant risks to privacy and societal security. Existing detection methods are predominantly passive, focusing on post-event analysis without preventing attacks. To address this, we propose an active defense method based on low-frequency perceptual perturbations to disrupt face swapping manipulation, reducing the performance and naturalness of generated content. Unlike prior approaches that used low-frequency perturbations to impact classification accuracy,our method directly targets the generative process of deepfake techniques. We combine frequency and spatial domain features to strengthen defenses. By introducing artifacts through low-frequency perturbations while preserving high-frequency details, we ensure the output remains visually plausible. Additionally, we design a complete architecture featuring an encoder, a perturbation generator, and a decoder, leveraging discrete wavelet transform (DWT) to extract low-frequency components and generate perturbations that disrupt facial manipulation models. Experiments on CelebA-HQ and LFW demonstrate significant reductions in face-swapping effectiveness, improved defense success rates, and preservation of visual quality.

Key findings

Experiments show significant reductions in face-swapping effectiveness and improved defense success rates across multiple deepfake generation models on both CelebA-HQ and LFW datasets. The approach maintains high visual quality in the perturbed images.

Approach

The authors propose a three-component architecture (encoder, perturbation generator, decoder) leveraging the Discrete Wavelet Transform (DWT). The encoder extracts low-frequency components, the generator creates perturbations targeting these components, and the decoder reconstructs the perturbed image. This process aims to disrupt the face-swapping generative model without significantly impacting visual quality.

Datasets

CelebA-HQ, LFW

Model(s)

The paper does not specify a particular pre-trained model for deepfake generation, but evaluates its approach against multiple models including SimSwap, InfoSwap, UniFace, E4S, and StarGan v2. The proposed method is an active defense mechanism, not a detection model.

Author countries

China

← Previous