The eyes may be the windows to the soul, but your face may provide law enforcement and security systems a more sure-fire way to identify you.
Researchers at the University of Sussex and London's Queen Mary and Westfield College have jointly developed a new system for facial recognition that uses moving images to track and identify faces, storing them in a database for later recall and matching.
Intended to handle relatively small - fewer than 100 - groups, the system runs on ordinary PCs and is intended for use in applications such as home-based security systems, office monitoring, and tracking of shopper preferences in supermarkets and other retail areas.
Jonathan Howell, the researcher leading the recognition effort at Sussex's School of Cognitive Science, says his project is designed to recognize faces from a variety of angles, particularly profiles, which is "more useful in a real-life environment." With views from different angles of a face, the system can easily identify people as they move. By contrast, most previous research has used a static camera, which limits the ability to identify faces, Howell says.
Howell's system also uses a small Radial Basis Function, a neural network that increases the speed at which it processes the images. "If you're looking at video sequences, you have to be able to finish processing in the 1/25th of a second between frames," he says. Other systems take several seconds per face.
Speed is also possible because the system needs to perform only a small number of calculations to process the images.
Howell, who expects to finish his PhD this summer, is hoping to secure another two-year grant to continue his work. Already, he sees near-term benefits such as better television research. But future iterations of this technology could help people solve the perplexities of interacting with VCRs and other household devices that are now becoming more computer-like. By adding a silicon chip, manufacturers inadvertently add to the complexity of the household as one device may have an interface that differs from that of another device.
But Howell imagines that the project's technology - with more development - could become part of an appliance-management system that serves as a hub for all these devices. A person would need only to learn one interface to interact with all appliances.
