New Facebook algorithm blocks facial recognition identification in videos
Facebook has developed a machine learning algorithm based on encoder-decoder architecture to protect individuals from being identified with biometric facial recognition in videos, writes Venture Beat. The de-identification system created by Facebook AI Research is one of the first to work with video, as other solutions developed in the past were designed for still images.
In August, for example, Tel Aviv-based startup D-ID introduced a new anonymization solution that blocks biometric features, personally identifiable information (PII) and license plates from video and still images. Earlier this year, the company won a Netexplo award for its artificial intelligence technology.
According to Venture Beat, Facebook’s AI solution does not need extensive training for each video it is applied to; it simply alters facial features to create a mask that prevents identification through facial recognition. The method tries to trick facial recognition into generating a different face.
“Face recognition can lead to loss of privacy and face replacement technology may be misused to create misleading videos,” reads the paper explaining how the system works. “Recent world events concerning the advances in, and abuse of face recognition technology invoke the need to understand methods that successfully deal with de-identification. Our contribution is the only one suitable for video, including live video, and presents quality that far surpasses the literature methods.”
According to Facebook AI Research engineer and Tel Aviv University professor Lior Wolf, the same autoencoder can be applied to “mask somebody’s, say, voice or online behavior or any other type of identifiable information that you want to remove,” he said in an interview with Venture Beat.
For now, there are no plans to integrate the software with Facebook applications, according to a Facebook spokesperson. The method will be showcased at the International Conference on Computer Vision (ICCV) this week in Seoul, South Korea.
In September, Facebook, the Partnership on AI, Microsoft, and academics from the U.S. and UK together launched a Deepfake Detection Challenge (DFDC) for AI developers and researchers to work on upgrading deepfake detection systems. Facebook contributed $10 million. CTO Mike Schroepfer announced the company was working on a database of deepfake videos made with paid actors for community use. Around that time, Facebook eliminated the facial recognition-based tag suggestion feature.
Just last week, Twitter announced it was considering new policies to address deepfakes to prevent the dissemination of false information in media, while Amazon announced it was joining the Deepfake Detection Challenge.