Facebook’s New, Advanced Facial Recognition

Facebook’s New, Advanced Facial Recognition

This post is also available in: heעברית (Hebrew)

Facebook developed a new facial recognition software, operating on a near-human level

12993596_m featureHuman beings, or at least our brains, are oiled facial recognition machines. Light, shadow, different angles, blurry, masked – our brain can effortlessly identify faces with high levels of precision. Facebook’s researchers recently developed a new software that can reach that high, almost human level of facial recognition capability.

This new development springs from a new approach to machine intelligence called “deep learning”, adopted last year by Facebook and its competitors. Deep-learning software means software that simulates the human brain’s neurons in an attempt to recognize patterns in large amounts of data. According to Facebook’s AI team, established last year, they managed to reach near-human levels of performance.

The new software’s name is DeepFace. According to the researchers it’s a “facial verification” program – identifying faces appearing in two pictures as the same face – rather than “facial recognition” – linking a face to a name. Researchers added, however, that in the future similar methods could be used to improve Facebook’s image-tagging suggestions.

IHLS – Israel Homeland Security

The software processes pictures in two stages. In the first stage it corrects the angle of the face, making the person appear to face forward. In the second stage it processes the face into a series of numeric values. If the numeric values of different photos are similar enough to each other, it’s probably the same face.

The program is still in research stages right now. Facebook, according to the researchers, has shown how feeding large amounts of information – in this case millions of social network user photographs – into a simulated neural network can significantly improve AI performance.

Source: MIT Technology Review