ExS-GAN: Synthesizing Anti-Forensics Images via Extra Supervised GAN

Area of Science:

• Digital Forensics
• Machine Learning
• Cybersecurity

Background:

• Digital forensics tools are crucial for data authenticity and integrity.
• Machine learning advancements pose new threats to existing forensics tools.
• Generative Adversarial Networks (GANs) present a significant anti-forensics challenge.

Purpose of the Study:

• To investigate the anti-forensics capabilities of GANs.
• To propose a novel GAN-based model for attacking digital forensics detectors.
• To enhance the development of countermeasures for digital forensics.

Main Methods:

• Proposed a novel Generative Adversarial Network (GAN) model named ExS-GAN.
• Incorporated an extra supervision system into the GAN architecture.
• Evaluated the anti-forensics performance on various manipulated images.

Main Results:

• The ExS-GAN model demonstrated high anti-forensics performance.
• The proposed method successfully attacked existing forensics detectors.
• Satisfying image quality was preserved during the anti-forensics attacks.
• An ablation study validated the effectiveness of the extra supervision system.

Conclusions:

• GANs pose a viable threat to digital forensics tools.
• The ExS-GAN model offers a robust method for anti-forensics research.
• Findings aid in developing more resilient digital forensics countermeasures.