Face Omron Ring: Proactive defense against face forgery with identity awareness

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Artificial Intelligence Generated Content (AIGC) models, particularly face forgery techniques, present significant security risks.
• The creation of deepfakes targeting public figures and persons-of-interest (POI) has led to widespread negative societal impacts.
• Existing defense mechanisms often lack proactive, identity-centric protection against sophisticated face forgery.

Purpose of the Study:

• To introduce a novel defense mechanism, the Face Omron Ring (FOR), for proactive protection against face forgery models.
• To develop a method that allows face forgery models to selectively refuse forging images of protected identities.
• To ensure that the protection of specific identities does not compromise the general forgery capabilities for unprotected individuals.

Main Methods:

• The proposed Face Omron Ring (FOR) is integrated into existing face forgery models.
• Experiments were conducted using four prominent face forgery models: StarGAN, AGGAN, AttGAN, and HiSD.
• Extensive testing was performed on large-scale datasets including CelebA, CelebA-HQ, and PubFig83.

Main Results:

• The FOR method successfully protected 5000 distinct identities with a 100% protection success rate, requiring approximately 100 images per identity.
• The proposed approach demonstrated high robustness against common image processing attacks like JPEG compression, cropping, noise addition, and blurring.
• FOR provides identity-centric protection without requiring pre-processing, enhancing scalability and security compared to existing methods.

Conclusions:

• The Face Omron Ring (FOR) offers an effective and scalable solution for proactive, identity-centric defense against face forgery.
• This method provides a pathway for responsible AIGC companies to regulate the misuse of face forgery technologies.
• FOR enhances security by enabling targeted protection of individuals while maintaining the functionality of AI models for other applications.