Memory-based Graphics: A New Technology For Visual Interfaces And Interactive Visualization

This project is based on a novel computer graphics technology for man/machine interfaces that relies on machine learning techniques, neural-like networks, and object recognition IU techniques. The key idea is to use a few "example" views of an object or person, to teach the computer to generate other views, under user control. WIth our technology we will develop an interactive (multimedia) interface that will enable the computer to communicate with the user with facial expressions and possibly synthesized speech. The technology advance is based on developments at M.I.T. of a new technique for learning from examples (the "HyperBF network"), partly under DARPA and ONR funding, and to its specific application to computer graphics and animation (Poggio and Brunelli, Patent Application No. 7-819767 and Poggio and Librande, MIT Case No. 5831S). The technology is closely related to work at MIT in Professor Poggio's group on object recognition, which combines classicial computer vision and networks for learning. ANTICIPATED BENEFITS: The main applications of the technology are to create new man/machine interfaces and enhance the images found in electronic documents, such as computer manuals and reference materials. Other applications that will benefit from this technology are virtual reaility simulations, interactive multimedia and very-high compression teleconferencing.