Boffins at Facebook have developed an algorithm that can recognise a person in a photo, even if their face is hidden from view.
The experimental algorithm, which was developed by Facebook’s artificial intelligence lab, looks for characteristics like hairstyle, clothing, body shape, pose and more.
New Scientist reports that the software mimics the human ability to recognise others without seeing their face.
The algorithm is understood to be 83pc accurate at picking people out from a crowd.
For example, the head of the artificial intelligence lab at Facebook, Yann LeCun, said that you could recognise Mark Zuckerberg very easily, because he always wears a grey t-shirt.
Accuracy of Facebook’s AI software
Facebook’s research team pulled almost 40,000 public photos from Flickr, some with the subject’s full face visible and others turned away.
They then ran them through a sophisticated neural network and the final algorithm was able to recognise people’s identities with 83pc accuracy.
The algorithm could soon power Facebook’s Moment’s app.
However, for people who value their privacy, the ability to be spotted in pictures via social media software is not necessarily welcome.
The Facebook Moments app, which automatically tags people in photos, has not been launched in Europe because of myriad EU rules on privacy.
Face recognition image, via Shutterstock
