Abstract

Classical techniques of designing image analyzers work well when the data in the field are similar to the data used in the design — e.g., in optical character recognition. There are many applications however in which it is very difficult or impossible to generalize adequately from the available design data by classical techniques [1]. Furthermore, requirements of real-time image processing often demand that we exploit uncertain data in making reliable decisions. Toward overcoming these difficulties in the design of real-time image analyzers we introduce a theory of dynamic belief. We refer to the image analyzers obtained with the aid of this theory as dynamic belief systems.

© 1989 Optical Society of America