(Credit: Matt Harnack/Facebook)
Facebook is working on artificial intelligence software called DeepFace that is capable of recognizing matching faces in images with nearly the same accuracy level as humans.
The social network’s DeepFace system uses a 3D modeling technique to detect faces, and crop and warp them so that they face front, a method known as frontalization.
The in-testing software is a facial verification system and differs from facial recognition in that it matches faces in large datasets, as opposed to assigning identity to faces. In essence, DeepFace can scan millions of photos, virtually rotate and correct the images, and find all matching faces.
Facebook’s DeepFace alignment system uses 2D and 3D facial modeling and deep learning to arrive at a final frontalized crop (g).
The sophisticated system was trained using a dataset of more than 4 million facial images of 4,000 people. Facebook’s method proved accurate 97.25 percent of the time, according to the company’s recently published paper, “DeepFace: Closing the Gap to Human-Level Performance in Face Verification.”
Though still in the research and development stages, Facebook’s proposed system purports to reduce the error of the current state of facial matching technologies by more than 25 percent.
Facebook’s AI Group will present its research at the Conference on Computer Vision and Pattern Recognition in June.
[via MIT Technology Review]
IBM to take Watson mobile with developer challenge
Bot or not? Try to tell a human poet from a computer
Tim Berners-Lee: 25 years on, the Web still needs work (Q&A)
Moov fitness tracker actually tells us how to fix ourselves
Adobe brings Revel photo-sync service to Android