Abstract

An electro-optic approach to the selection of the closest match between an object described by a feature vector and a set of known feature vectors is described. The problem is formulated as parallel vector-matrix multipliers and a summing operation of the form: Typical problems usually do not call for an exceptionally rapid update of the stored information, while the rapid sorting operation is generally a strong requirement. These conditions can be satisfied with currently available technology. An implementation is described in which feature vector data is stored on a liquid crystal device utilized as a window addressable memory. A description of breadboard hardware is given. Preliminary test results support the overall feasibility of the approach.

© 1985 Optical Society of America