Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Active Fake: DeepFake Camouflage

Pu Sun, Honggang Qi, Yuezun Li

TL;DR

This work introduces a new framework for creating DeepFake camouflage that generates blending inconsistencies while ensuring imperceptibility, effectiveness, and transferability, and crafts imperceptible yet effective inconsistencies to mislead forensic detectors.

Abstract

DeepFake technology has gained significant attention due to its ability to manipulate facial attributes with high realism, raising serious societal concerns. Face-Swap DeepFake is the most harmful among these techniques, which fabricates behaviors by swapping original faces with synthesized ones. Existing forensic methods, primarily based on Deep Neural Networks (DNNs), effectively expose these manipulations and have become important authenticity indicators. However, these methods mainly concentrate on capturing the blending inconsistency in DeepFake faces, raising a new security issue, termed Active Fake, emerges when individuals intentionally create blending inconsistency in their authentic videos to evade responsibility. This tactic is called DeepFake Camouflage. To achieve this, we introduce a new framework for creating DeepFake camouflage that generates blending inconsistencies while ensuring imperceptibility, effectiveness, and transferability. This framework, optimized via an adversarial learning strategy, crafts imperceptible yet effective inconsistencies to mislead forensic detectors. Extensive experiments demonstrate the effectiveness and robustness of our method, highlighting the need for further research in active fake detection.

Active Fake: DeepFake Camouflage

TL;DR

This work introduces a new framework for creating DeepFake camouflage that generates blending inconsistencies while ensuring imperceptibility, effectiveness, and transferability, and crafts imperceptible yet effective inconsistencies to mislead forensic detectors.

Abstract

DeepFake technology has gained significant attention due to its ability to manipulate facial attributes with high realism, raising serious societal concerns. Face-Swap DeepFake is the most harmful among these techniques, which fabricates behaviors by swapping original faces with synthesized ones. Existing forensic methods, primarily based on Deep Neural Networks (DNNs), effectively expose these manipulations and have become important authenticity indicators. However, these methods mainly concentrate on capturing the blending inconsistency in DeepFake faces, raising a new security issue, termed Active Fake, emerges when individuals intentionally create blending inconsistency in their authentic videos to evade responsibility. This tactic is called DeepFake Camouflage. To achieve this, we introduce a new framework for creating DeepFake camouflage that generates blending inconsistencies while ensuring imperceptibility, effectiveness, and transferability. This framework, optimized via an adversarial learning strategy, crafts imperceptible yet effective inconsistencies to mislead forensic detectors. Extensive experiments demonstrate the effectiveness and robustness of our method, highlighting the need for further research in active fake detection.
Paper Structure (16 sections, 9 equations, 8 figures, 9 tables)

This paper contains 16 sections, 9 equations, 8 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. DeepFake
  4. DeepFake Detection
  5. Evading DeepFake Detection
  6. Active Fake: DeepFake Camouflage
  7. Inspiration and Preliminary Analysis
  8. Problem Setup & Overview
  9. Camouflage GAN
  10. Loss functions and Training
  11. Experiments
  12. Experimental Setups
  13. Results
  14. Ablation Study
  15. Further Analysis
  16. ...and 1 more sections

Figures (8)

  • Figure 1: Pipeline of creating a Face-Swap DeepFake face. Firstly, the central face area is cropped out from the original face image, which is used to synthesize a target face. Secondly, this face is blended back to the original face image using a specific mask.
  • Figure 2: Overview of DeepFake camouflage.
  • Figure 3: Grad-CAM maps on different images. Row (a) & (b): DeepFake images and their Grad-CAM maps. Row(c) & (d):Real clean images and their Grad-CAM maps. Row(e) & (f): Real images with handcrafted artifacts and their Grad-CAM maps.
  • Figure 4: Training pipeline of Camouflage GAN. The inference pipeline is in the gray background area. The green (real) and blue (fake) arrows represent the expected network output when the corresponding images are input in the training phase. The dashed arrows indicate that, when optimizing the configuration generator, the goal output of the visual discriminator taking camouflaged images as input should be Real; When optimizing the visual discriminator, the goal output of the visual discriminator taking camouflaged images as input should be Fake; See Sec. \ref{['sec:CamGAN']} for details.
  • Figure 5: Qualitative Results. We enlarge the areas within the green box for better view. Row (a) represents real clean images. Row (b) represents images camouflaged by our method. Row (c)-(f) represent images attacked by CW, Jitter, PGD, and Pixle, respctively. Note that none of these attack methods could compare with ours in terms of attack success rate. Zoom in to see the details.
  • ...and 3 more figures