Originally published 6 August 2015 on MIT Technology Review – Arab Edition.
Researchers at KAUST are developing an algorithm that can accurately recognize faces.
Recognizing faces is something we do every day at work, at home, or in public areas. People can recognize familiar faces quite well even with significant changes in the lighting, in the orientation of the head, when someone is wearing sunglasses, when a hairstyle has changed, when facial hair increases or decreases, when makeup is applied, at different distances and more, says researcher Matthew Turk from University of California. Automatic face recognition by computers on the other hand, is still a work in progress.
While it achieves good results in well-controlled environments, performance is rather poor in less constrained ones (also referred to as “in the wild” recognition) due to these variations. “Recognizing the face of an uncooperative person, or someone in surveillance video is much more difficult than doing so in a controlled setting where the person wants to be recognized (in order to log into a computer or gain access to a secure area, for example),” says Turk.
An algorithm to recognize faces
Researchers from the King Abdullah University of Science and Technology (KAUST) came up with a method to accurately simplify the matrix of faces by minimizing the maximum distance of a pair of faces from the same person and maximizing the minimum distance between a pair of faces from different people using non-negative matrix factorization (NMF), a state-of-the-art feature extraction algorithm.
Read more at MIT Technology Review – Arab Edition, August 2015.