This technology is a light-based authentication platform that uses imperceptible, tamper-resistant signatures during live recordings to verify authenticity and prevent deepfake manipulation.
Generative adversarial networks are increasingly being used to generate photo-realistic deepfakes that spread misinformation and damage reputations. Current deepfake detection algorithms are vulnerable to adversarial attacks, while digital watermarking and signing methods require cooperation across hardware and software ecosystems and can be removed by generative AI. There are currently no robust, secure, and hardware-independent methods to verify the authenticity of live-recorded videos and protect the integrity of visual media.
This technology, termed VeriLight, is a light-based authentication platform that embeds imperceptible, tamper-resistant signatures into videos during live speech recordings. The platform captures a speaker’s facial identity and motion features, compresses them into unique digital hashes, and uses them to generate a signature encoded as subtle light modulation projected near the speaker. Even though they are invisible to the viewer, the light patterns can later be decoded from video recordings to verify the source and detect falsification of identity or facial movements.
This technology has been validated across a range of recording angles and distances, demonstrating robustness and security compared to existing watermarking or metadata-based methods.
IR CU26073
Licensing Contact: Dovina Qu
