Abstract Representation For Model-Based Computer Vision

摘要

The current work presented is research into a general and flexible representationtechnique for model based computer vision. This abstract representation integrates various sources of knowledge within model based vision: functional, geometric, and relational; and provides a representation to express image data, image extracted parameters, and model parameters. The abstract representation is based upon the notion of feature structures as derived from linguistic applications of unification grammars and the manipulation technique of unification. The representation of feature structures and the manipulation technique of unification are combined into a lattice structure that is partially ordered by the subsumption relationship. These aspects are explored as the backbone of the abstract representation of the unification grammar application to model based computer vision. Promising aspects of accomplishing parallel unification and unifying solution search lattices in parallel are discussed. An evaluation of the major results of this work and a discussion on areas for future research conclude this dissertation.