Researchers have developed a new facial recognition algorithm that is adept at scanning faces positioned in a wide variety of angles, according to an article by Mary-Ann Russon for the International Business Times. The new algorithm is said to be a major step forward from the pioneering work of Michael Jones and Paul Viola that set the standard for the technology back in 2001.
The new algorithm is the product of computer scientists associated with Stanford University (Mohammad Saberian and Sachin Farfade) and Yahoo Labs (Li-Jia Li). They built a database of 200,000 faces posed in a variety of angles, plus 20 million faceless objects, and set a computer on the task of looking for faces in 128-image batches. That went on in 50,000 iterations, as a process of deep machine learning. The piece of artificial intelligence that emerged from that feat was dubbed the Deep Dense Face Detector, and the researchers say that they plan to further augment the algorithm to improve detection of rotated and partially obscured faces.
The new algorithm, if it lives up to its hype, could prove to be a powerful force in a variety of applications. Facial recognition technology is increasingly being used in security deployments, particularly with respect to surveillance, and the ability to identify faces at a variety of angles and possibly even faces that are partially obscured, could give security agents a huge leg up. It could also help to improve the technology in more everyday applications, such as identifying customers in retail settings, or helping to more quickly unlock mobile devices.
February 20, 2015 – by Alex Perala