Summary:
Researchers in the UCLA Department of Electrical and Computer Engineering have developed a multi-layer security framework to verify deepfake imagery data.
Background:
The exponential improvements in generative AI pose serious implication to the rise of synthetic media or “deepfakes”. Soon, deepfake images and videos will be completely indistinguishable from real ones and the AI models used to generate them will become broadly available. While beneficial uses of deepfakes for educational or entertainment exist, bad actors could use them to harm individuals or business, influence national elections, and even produce non-consensual pornographic materials. To mitigate the risks posed by deepfakes, there is an urgent need to develop security and algorithmic protocols to label real data and detect fake data.
Innovation:
Researchers in the Department of Electrical and Computer Engineering and Department of Computer Science at UCLA have developed a new cross-layer security system to ensure photography and digital imagery can be verified as real. They achieve this using an image signal processing algorithm that provides end-to-end data provenance, ensuring image authenticity. They further developed a method to reduce future threats of physical print out attacks. In addition, they describe a novel file format accompanied by a third-party verification tool to improve image verification. Taken together these innovations provide significant improvements to image security and authenticity which is essential to prevent threats posed by deepfake data.
Potential Applications:
• Image authentication
• Deepfake identification and prevention
• Data security system
• Social media content moderation
Advantages:
• Compatible with image or video data
• System can be deployed on mobile devices
• File system embeds authenticity natively to data file
Development-To-Date:
Researchers have developed an Android-based prototype for determining image provenance.
Reference:
UCLA Case No. 2024-270