Object Classification through Probabilistic Common Sense Knowledge Reasoning

Abstract

This thesis presents a manner for object classification by the use of semantic knowledge and probabilistic reasoning with such knowledge. An ontology of object classes and their context and properties is represented as a Markov Logic Network, which is a method of unifying first-order logic with probabilistic reasoning, developed recently. For each scene, the ontology is combined with symbolic observations of objects observed in the scene. Probabilistic inference is then used to infer the class or a superclass of those objects.