Facebook DeepFace: Facial Recognition at a new level

Though facial recognition is a modern technology that aims at the replacement of normal typed passwords, it is not quite in vogue. The reason is also genuine, ‘It is not so accurate’. So to overcome this issue, Facebook is further developing its facial verification technology to make it nearly as accurate as the human eye. Very soon, the social networking giant may be able to recognize your face just as precisely as you can!

In a recent post, Facebook explained how the conventional pipeline of facial recognition goes. It explains that the process consists of four stages:  detection -> alignment -> representation -> classification, where both the alignment step and the representation step are revisited by employing explicit 3D face modeling in order to apply a piece-wise affine transformation, and derive a face representation from a nine-layer deep neural network involving more than 120 million parameters using several locally connected layers without weight sharing, rather than the standard convolutional layers. So, it claims to have trained it on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4,000 identities, where each identity has an average of over a thousand samples.


In simple words, first, it corrects the subject’s face so that the person is facing forward in the image. It uses a 3D model of an “average” forward-looking face to nail down this angle. Then, the software uses a method known as deep learning, which means it simulates a neural network that can create a numerical description of reoriented face. If the software finds similar enough facial descriptions for two different images, it concludes that they must be the same face.

This technology currently under research can prove to be 97.25% accurate in unconstrained environments which the company says is 25 percent more accurate than it was before. The DeepFace software, as Facebook calls it for now, will likely prevent the website from mistakenly tagging you in photos as a friend that may have similar facial features.

DeepFace, however, is just a research project for now, according to MIT, but researchers will be presenting the technology at the IEEE Conference on Computer Vision and Pattern Recognition in June.

Read More: Facebook Publication / MIT Technology Review


comments powered by Disqus