Browse Prior Art Database

Constraint-Networks: Modeling and-Inferring Objects Locations by Constraints

IP.com Disclosure Number: IPCOM000128622D
Original Publication Date: 1978-Dec-31
Included in the Prior Art Database: 2005-Sep-16
Document File: 13 page(s) / 48K

Publishing Venue

Software Patent Institute

Related People

Daniel M. Russell: AUTHOR [+3]

Abstract

Relationships between objects in the real world are constrained by many physical and functional considerations. This paper presents a formalism called Constraint Networks which allows such constraints to be represented and used to make infer-ences about object locations in images. Constraint Networks are used in a system which accepts information about geometric rela-tionships between structures in images and than uses these constraints to guide search for these structures. The system has been used successfully to infer rib positions in a chest X-ray and to locate aeration tanks and new construction sites in aerial images.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 8% of the total text.

Page 1 of 13

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

Constraint-Networks: Modeling and-Inferring Objects Locations by Constraints

Daniel M. Russell Computer Science Department The University of Rochester Rochester, NY 14627

TR38

August 1978

Abstract

Relationships between objects in the real world are constrained by many physical and functional considerations. This paper presents a formalism called Constraint Networks which allows such constraints to be represented and used to make infer-ences about object locations in images. Constraint Networks are used in a system which accepts information about geometric rela- tionships between structures in images and than uses these constraints to guide search for these structures. The system has been used successfully to infer rib positions in a chest X-ray and to locate aeration tanks and new construction sites in aerial images.

The preparation of this paper was supported in part by the Defense Advanced Research Projects Agency, monitored by the ONR under Contract `do. "d000I4-78-C-0164

I. Introduction

Consider the following situation: You are flying at 3000 feet somewhere close to home and wish to find your neighbor's house; you might find yourself examining the scene below in the following fashion: "Well, I know I live close to the river. Upstream from the end of my street is .a park, so if I find the park I can find my~house. And since my neighbor lives north of my house I can just look a little north of my house...". This impromptu strategy for neighbor's-house-finding can be viewed, as applying a successive series of constraints to the aerial image, thus removing areas of the scene from further consideration. "My house is close to a river and down from the park" and "qty neighbor's house is close to mine" are both facts which can be used to limit the search space for a feature to a reasonable size. The first constrains the sub-image to be scrutinized to that part of the total image which is both close to the river and at a particular orientation to the river and park, while the second implies a constrained sub-image which is close to your house. This process of successive reduction of the search space by repeated limitation of the space is extremely useful in approaching a computer vision task. Currently, many systems simply use the technology approach to vision; i.e. apply an operator over the whole of an image, and then use the results from this massive "search" for further analysis. If instead we could limit - or, as we will say, constrain - our focus of attention to a much smaller area by the intelligent use of some facts and inferences drawn from them, we could then make considerable savings in an analysis of any scene.

Our goal, then, is to maximize the use of facts we already know about a given scene to tell us where to look for objects we are interested in. Further, we would also like to be able to use any

University of Rochester Page 1 Dec 31, 1978

Page 2 of 13

Con...