Revolutionary AI ‘Chameleon’ Protects Personal Photos from Facial Recognition

Artificial intelligence (AI) can help protect your personal photos from facial recognition software and fraud. A study from Georgia Tech introduced an AI model called “Chameleon.” This model creates a personalized privacy protection mask for photos that confuses facial recognition systems, making them misidentify the person in the image.

Ling Liu, a professor at Georgia Tech, led the research. She highlighted that tools like Chameleon can support responsible AI use and promote innovation. Facial recognition technology is widespread in daily life, found in police cameras and smartphones. However, unauthorized scans can lead to issues like fraud or unwanted advertising.

Chameleon has three main features. First, it uses cross-image optimization, which means it generates one mask for each user instead of creating a new one for every image. This method speeds up protection and conserves computing resources.

Second, Chameleon employs perceptibility optimization. This feature ensures the visual quality of protected images remains high without needing manual adjustments.

The third feature enhances the mask’s strength to counter unfamiliar facial recognition systems. It does this through ensemble learning, a technique that combines predictions from multiple models to improve accuracy.

The researchers aim to extend Chameleon’s techniques to protect images from being exploited for AI training without consent. Tiansheng Huang, a Georgia Tech doctoral student, emphasized the importance of protecting personal images from unauthorized use.