Facebook’s DeepFace Can Recognize Your Face in the Crowd with Disturbing Accuracy

Facebook’s database of personal information about its users grows daily in astronomical leaps, and it would be naive to assume the company isn't doing anything with that huge load of information. Last week, the company published an update revealing its potent facial recognition software, DeepFace, which allows Facebook servers to identify users in photographs with creepy accuracy.

The software’s name stems from an artificial intelligence project known as Deep Learning, the system that involves collecting massive amounts of data and analyzing them through a complicated algorithm, training it to recognize patterns in sounds, images and other media.

In this case, the company sits atop a treasure cave of user data of voluntarily uploaded personal pictures, some of them being selfies. According to Facebook, DeepFace identified 97.25% of 4 million facial images by 4000 users, matching photographs to individuals regardless of position of the face, shadow and other complicating factors. The current success rate eliminates the possibility of errors in identifying individuals almost completely.

If you wonder how DeepFace works, here’s a brief breakdown of its algorithm. For example, the AI is given an image of a woman facing left. The algorithm maps her features in 3-D and turns her head forward virtually to analyze that forward facing face and produces a “numeric map” of her face to be able to search for more patterns and find the perfect match in the database. Once the algorithm finds two patterns that indicate very similar models, it signals it has found a match.

While Facebook is probably going to apply this technology to make automatic tagging of images quicker and more streamlined, it is also possible the company could dwell more on the additional purpose of the technology during the Conference On Computer Vision And Pattern Recognition in June.