Dynamic structuring of belief networks in a hierarchical perceptual organization

Proposes a hierarchical perceptual organization (HPO) for doing scene interpretation. The HPO consists of three parts; (1) an input data base and the static models, (2) the active models, and (3) the rule-based decision part. Static models are composed of a set of physical models and a set of relational models stored as a general knowledge-base. The active models are belief networks which are dynamically created and grown from the static models using the rules in the decision part. The outputs of the model are hypotheses on the objects in the scene. The rule-based decisions control the forming of the structure of the belief networks. The main focus of the paper is on the dynamic structuring of the belief networks under the influence of the rule-based decision part