Considering multiple surface hypotheses in a Bayesian hierarchy

Steven M. Lavalle, Seth A. Hutchinson

Research output: Contribution to journalConference article

Abstract

In this paper, we present a probabilistic approach to segmentation which maintains a set of competing, plausible segmentation hypotheses. This is in contrast to previous approaches, in which probabilistic methods are used to converge to a single segmentation. The benefit of our approach is that belief values associated with segmentation hypotheses can be used to guide the recognition process, and, the recognition process can, in turn, exert influence on the belief values associated with segmentation hypotheses in the network. In this way, segmentation and recognition can be coupled together to achieve a combination of expectation driven segmentation and data-driven recognition. Our algorithms are based on the formalism of Bayesian belief networks. By storing segmentation hypotheses in a tree structured network, we are able to limit the storage demands associated with maintaining the competing hypotheses. We have also introduced an implicit representation for segmentation hypotheses (without this implicit representation, we would be forced to enumerate the power set of region groupings). Likelihood measures are used both to control the expansion of the hypothesis tree and to evaluate belief in hypotheses. Local likelihood measures are used during an expansion phase, in which leaf nodes are refined into more specific hypotheses. Global likelihood measures are applied during an evaluation phase. The global likelihood measures are derived by fitting quadric surfaces to the range data. By using this expand and evaluate approach guided by a measure of entropy defined on the leaves of the tree, we are able to limit the application of costly numerical fitting algorithms to a small number of nodes in the tree.

Original languageEnglish (US)
Pages (from-to)2-15
Number of pages14
JournalProceedings of SPIE - The International Society for Optical Engineering
Volume1569
DOIs
StatePublished - Oct 1 1991
EventStochastic and Neural Methods in Signal Processing, Image Processing, and Computer Vision 1991 - San Diego, United States
Duration: Jul 21 1991 → …

Fingerprint

hierarchies
Segmentation
Bayesian networks
Entropy
Likelihood
leaves
belief networks
Leaves
Bayesian Belief Networks
Local Likelihood
Hierarchy
Power set
expansion
Evaluate
Probabilistic Methods
Probabilistic Approach
Quadric
Vertex of a graph
Data-driven
Grouping

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics
  • Computer Science Applications
  • Applied Mathematics
  • Electrical and Electronic Engineering

Cite this

Considering multiple surface hypotheses in a Bayesian hierarchy. / Lavalle, Steven M.; Hutchinson, Seth A.

In: Proceedings of SPIE - The International Society for Optical Engineering, Vol. 1569, 01.10.1991, p. 2-15.

Research output: Contribution to journalConference article

@article{b93aa2c3074341af8f78ec767eccb1a5,
title = "Considering multiple surface hypotheses in a Bayesian hierarchy",
abstract = "In this paper, we present a probabilistic approach to segmentation which maintains a set of competing, plausible segmentation hypotheses. This is in contrast to previous approaches, in which probabilistic methods are used to converge to a single segmentation. The benefit of our approach is that belief values associated with segmentation hypotheses can be used to guide the recognition process, and, the recognition process can, in turn, exert influence on the belief values associated with segmentation hypotheses in the network. In this way, segmentation and recognition can be coupled together to achieve a combination of expectation driven segmentation and data-driven recognition. Our algorithms are based on the formalism of Bayesian belief networks. By storing segmentation hypotheses in a tree structured network, we are able to limit the storage demands associated with maintaining the competing hypotheses. We have also introduced an implicit representation for segmentation hypotheses (without this implicit representation, we would be forced to enumerate the power set of region groupings). Likelihood measures are used both to control the expansion of the hypothesis tree and to evaluate belief in hypotheses. Local likelihood measures are used during an expansion phase, in which leaf nodes are refined into more specific hypotheses. Global likelihood measures are applied during an evaluation phase. The global likelihood measures are derived by fitting quadric surfaces to the range data. By using this expand and evaluate approach guided by a measure of entropy defined on the leaves of the tree, we are able to limit the application of costly numerical fitting algorithms to a small number of nodes in the tree.",
author = "Lavalle, {Steven M.} and Hutchinson, {Seth A.}",
year = "1991",
month = "10",
day = "1",
doi = "10.1117/12.48362",
language = "English (US)",
volume = "1569",
pages = "2--15",
journal = "Proceedings of SPIE - The International Society for Optical Engineering",
issn = "0277-786X",
publisher = "SPIE",

}

TY - JOUR

T1 - Considering multiple surface hypotheses in a Bayesian hierarchy

AU - Lavalle, Steven M.

AU - Hutchinson, Seth A.

PY - 1991/10/1

Y1 - 1991/10/1

N2 - In this paper, we present a probabilistic approach to segmentation which maintains a set of competing, plausible segmentation hypotheses. This is in contrast to previous approaches, in which probabilistic methods are used to converge to a single segmentation. The benefit of our approach is that belief values associated with segmentation hypotheses can be used to guide the recognition process, and, the recognition process can, in turn, exert influence on the belief values associated with segmentation hypotheses in the network. In this way, segmentation and recognition can be coupled together to achieve a combination of expectation driven segmentation and data-driven recognition. Our algorithms are based on the formalism of Bayesian belief networks. By storing segmentation hypotheses in a tree structured network, we are able to limit the storage demands associated with maintaining the competing hypotheses. We have also introduced an implicit representation for segmentation hypotheses (without this implicit representation, we would be forced to enumerate the power set of region groupings). Likelihood measures are used both to control the expansion of the hypothesis tree and to evaluate belief in hypotheses. Local likelihood measures are used during an expansion phase, in which leaf nodes are refined into more specific hypotheses. Global likelihood measures are applied during an evaluation phase. The global likelihood measures are derived by fitting quadric surfaces to the range data. By using this expand and evaluate approach guided by a measure of entropy defined on the leaves of the tree, we are able to limit the application of costly numerical fitting algorithms to a small number of nodes in the tree.

AB - In this paper, we present a probabilistic approach to segmentation which maintains a set of competing, plausible segmentation hypotheses. This is in contrast to previous approaches, in which probabilistic methods are used to converge to a single segmentation. The benefit of our approach is that belief values associated with segmentation hypotheses can be used to guide the recognition process, and, the recognition process can, in turn, exert influence on the belief values associated with segmentation hypotheses in the network. In this way, segmentation and recognition can be coupled together to achieve a combination of expectation driven segmentation and data-driven recognition. Our algorithms are based on the formalism of Bayesian belief networks. By storing segmentation hypotheses in a tree structured network, we are able to limit the storage demands associated with maintaining the competing hypotheses. We have also introduced an implicit representation for segmentation hypotheses (without this implicit representation, we would be forced to enumerate the power set of region groupings). Likelihood measures are used both to control the expansion of the hypothesis tree and to evaluate belief in hypotheses. Local likelihood measures are used during an expansion phase, in which leaf nodes are refined into more specific hypotheses. Global likelihood measures are applied during an evaluation phase. The global likelihood measures are derived by fitting quadric surfaces to the range data. By using this expand and evaluate approach guided by a measure of entropy defined on the leaves of the tree, we are able to limit the application of costly numerical fitting algorithms to a small number of nodes in the tree.

UR - http://www.scopus.com/inward/record.url?scp=84975671888&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84975671888&partnerID=8YFLogxK

U2 - 10.1117/12.48362

DO - 10.1117/12.48362

M3 - Conference article

AN - SCOPUS:84975671888

VL - 1569

SP - 2

EP - 15

JO - Proceedings of SPIE - The International Society for Optical Engineering

JF - Proceedings of SPIE - The International Society for Optical Engineering

SN - 0277-786X

ER -