Facebook has developed the next level of facial recognition software that is so clever, it can identify you even if your face is obscured. If you were paranoid about being auto-tagged in pictures before, Facebook’s new recognition capabilities won’t do anything to allay those fears. This new algorithm removes any residual layers of privacy a user would have from photographing themselves from the neck down, or covering their face. The AI behind the development seems human-like its ability to identify a friend from the back of their head.
Instead of relying on facial biometrics, Facebook’s newly developed algorithm uses other cues such as hairstyles, outfits, body type and posture to differentiate people in photographs. The final version of the algorithm which was presented at the Computer Vision and Pattern Recognition conference in Boston has an 83% accuracy.
According to Yann LeCun, who is in charge of Facebook’s artificial intelligence department, “There are a lot of cues we use. People have characteristic aspects, even if you look at them from the back.” He points out, “You can recognise Mark Zuckerberg very easily, because he always wears a gray T-shirt.”
This development comes on the heels of last week’s release of Facebook’s new photo app, Moments, which automatically curates event albums from photographs on your mobile device. The app is designed to identify and auto-tag any of your Facebook friends it can find in those photos. Moments is meeting resistance in Europe over privacy concerns. The app won’t be available in Europe until it gives users the chance to opt-out of being identified by facial recognition.
In contrast, the recognition algorithm could also be used to combat privacy abuses. In theory, it could aggregate all the photos that you appear in and notify you when a new image is released onto the Internet.
Source: New Scientist