A fundamental goal of computer vision is the development of systems capable of carrying out scene interpretation while taking into account all the available knowledge. In this report, we have focused on how the interpretation task may be aided by expected-scene information which, in most cases, would not be in registration with the perceived scene. In this report, we describe PSEIKI, a framework for expectation-driven interpretation of image data. PSEIKI builds abstraction hierarchies in image data using, for cues, supplied abstraction hierarchies in a scene expectation map. Hypothesized abstractions in the image data are geometrically compared with the known abstractions in the expected scene; the metrics used for these comparisons translate into belief values. The Dempster-Shafer formalism is used to accumulate beliefs for the synthesized abstractions in the image data. For accumulating belief values, a computationally efficient variation of Dempster’s rule of combination is developed to enable the system to deal with the overwhelming amount of information present in most images. This variation of Dempster’s rule allows the reasoning process to be embedded into the abstraction hierarchy by allowing for the propagation of belief values between elements at different levels of abstraction. The system has been implemented as a 2- panel, 5-level blackboard in OPS 83. This report also discusses the control aspects of the blackboard, achieved via a distributed monitor using the OPS83 demons and a scheduler. Various knowledge sources for forming groupings in the image data and for labeling such groupings with abstractions from the scene expectation map are also discussed.
Date of this Version