A graphical model framework for image segmentation

Research output: Chapter in Book/Report/Conference proceedingChapter

1 Citation (Scopus)

Abstract

Graphical models are probabilistic models defined in terms of graphs. The intuitive and compact graph representation and its ability to model complex probabilistic systems make graphical models a powerful modeling tool in various research areas. In this paper we introduce a graphical model framework for image segmentation based on the integration of Markov random fields (MRFs) and deformable models. A graphical model is constructed to represent the relationship of the observed image pixels, the true region labels and the underlying object contour. We then formulate the problem of image segmentation as the one of joint region-contour inference and learning in the graphical model. The graphical model representation allows us to use an approximate structured variational inference technique to solve this otherwise intractable joint inference problem. Using this technique, the MAP solution to the original model is obtained by finding the MAP solutions of two simpler models, an extended MRF model and a probabilistic deformable model, iteratively and incrementally. In the extended MRF model, the true region labels are estimated using the BP algorithm in a band area around the estimated contour from the probabilistic deformable model, and the result in turn guides the probabilistic deformable model to an improved estimation of the contour. Finally, we generalize our method from 2D to 3D. Experimental results on both synthetic and real images, in both 2D and 3D, show that our new hybrid method outperforms both the MRF-based and the deformable model-based methods using onlyhomogeneous constraints.

Original languageEnglish (US)
Title of host publicationApplied Graph Theory in Computer Vision and Pattern Recognition
EditorsAbraham Kandel, Horst Bunke, Mark Last
Pages43-63
Number of pages21
DOIs
StatePublished - Apr 19 2007

Publication series

NameStudies in Computational Intelligence
Volume52
ISSN (Print)1860-949X

Fingerprint

Image segmentation
Labels
Pixels

All Science Journal Classification (ASJC) codes

  • Artificial Intelligence

Cite this

Huang, R., Pavlovic, V., & Metaxas, D. (2007). A graphical model framework for image segmentation. In A. Kandel, H. Bunke, & M. Last (Eds.), Applied Graph Theory in Computer Vision and Pattern Recognition (pp. 43-63). (Studies in Computational Intelligence; Vol. 52). https://doi.org/10.1007/978-3-540-68020-8_2
Huang, Rui ; Pavlovic, Vladimir ; Metaxas, Dimitri. / A graphical model framework for image segmentation. Applied Graph Theory in Computer Vision and Pattern Recognition. editor / Abraham Kandel ; Horst Bunke ; Mark Last. 2007. pp. 43-63 (Studies in Computational Intelligence).
@inbook{77f2a9d61dae4a669fcb3d3a82ec8b59,
title = "A graphical model framework for image segmentation",
abstract = "Graphical models are probabilistic models defined in terms of graphs. The intuitive and compact graph representation and its ability to model complex probabilistic systems make graphical models a powerful modeling tool in various research areas. In this paper we introduce a graphical model framework for image segmentation based on the integration of Markov random fields (MRFs) and deformable models. A graphical model is constructed to represent the relationship of the observed image pixels, the true region labels and the underlying object contour. We then formulate the problem of image segmentation as the one of joint region-contour inference and learning in the graphical model. The graphical model representation allows us to use an approximate structured variational inference technique to solve this otherwise intractable joint inference problem. Using this technique, the MAP solution to the original model is obtained by finding the MAP solutions of two simpler models, an extended MRF model and a probabilistic deformable model, iteratively and incrementally. In the extended MRF model, the true region labels are estimated using the BP algorithm in a band area around the estimated contour from the probabilistic deformable model, and the result in turn guides the probabilistic deformable model to an improved estimation of the contour. Finally, we generalize our method from 2D to 3D. Experimental results on both synthetic and real images, in both 2D and 3D, show that our new hybrid method outperforms both the MRF-based and the deformable model-based methods using onlyhomogeneous constraints.",
author = "Rui Huang and Vladimir Pavlovic and Dimitri Metaxas",
year = "2007",
month = "4",
day = "19",
doi = "10.1007/978-3-540-68020-8_2",
language = "English (US)",
isbn = "3540680195",
series = "Studies in Computational Intelligence",
pages = "43--63",
editor = "Abraham Kandel and Horst Bunke and Mark Last",
booktitle = "Applied Graph Theory in Computer Vision and Pattern Recognition",

}

Huang, R, Pavlovic, V & Metaxas, D 2007, A graphical model framework for image segmentation. in A Kandel, H Bunke & M Last (eds), Applied Graph Theory in Computer Vision and Pattern Recognition. Studies in Computational Intelligence, vol. 52, pp. 43-63. https://doi.org/10.1007/978-3-540-68020-8_2

A graphical model framework for image segmentation. / Huang, Rui; Pavlovic, Vladimir; Metaxas, Dimitri.

Applied Graph Theory in Computer Vision and Pattern Recognition. ed. / Abraham Kandel; Horst Bunke; Mark Last. 2007. p. 43-63 (Studies in Computational Intelligence; Vol. 52).

Research output: Chapter in Book/Report/Conference proceedingChapter

TY - CHAP

T1 - A graphical model framework for image segmentation

AU - Huang, Rui

AU - Pavlovic, Vladimir

AU - Metaxas, Dimitri

PY - 2007/4/19

Y1 - 2007/4/19

N2 - Graphical models are probabilistic models defined in terms of graphs. The intuitive and compact graph representation and its ability to model complex probabilistic systems make graphical models a powerful modeling tool in various research areas. In this paper we introduce a graphical model framework for image segmentation based on the integration of Markov random fields (MRFs) and deformable models. A graphical model is constructed to represent the relationship of the observed image pixels, the true region labels and the underlying object contour. We then formulate the problem of image segmentation as the one of joint region-contour inference and learning in the graphical model. The graphical model representation allows us to use an approximate structured variational inference technique to solve this otherwise intractable joint inference problem. Using this technique, the MAP solution to the original model is obtained by finding the MAP solutions of two simpler models, an extended MRF model and a probabilistic deformable model, iteratively and incrementally. In the extended MRF model, the true region labels are estimated using the BP algorithm in a band area around the estimated contour from the probabilistic deformable model, and the result in turn guides the probabilistic deformable model to an improved estimation of the contour. Finally, we generalize our method from 2D to 3D. Experimental results on both synthetic and real images, in both 2D and 3D, show that our new hybrid method outperforms both the MRF-based and the deformable model-based methods using onlyhomogeneous constraints.

AB - Graphical models are probabilistic models defined in terms of graphs. The intuitive and compact graph representation and its ability to model complex probabilistic systems make graphical models a powerful modeling tool in various research areas. In this paper we introduce a graphical model framework for image segmentation based on the integration of Markov random fields (MRFs) and deformable models. A graphical model is constructed to represent the relationship of the observed image pixels, the true region labels and the underlying object contour. We then formulate the problem of image segmentation as the one of joint region-contour inference and learning in the graphical model. The graphical model representation allows us to use an approximate structured variational inference technique to solve this otherwise intractable joint inference problem. Using this technique, the MAP solution to the original model is obtained by finding the MAP solutions of two simpler models, an extended MRF model and a probabilistic deformable model, iteratively and incrementally. In the extended MRF model, the true region labels are estimated using the BP algorithm in a band area around the estimated contour from the probabilistic deformable model, and the result in turn guides the probabilistic deformable model to an improved estimation of the contour. Finally, we generalize our method from 2D to 3D. Experimental results on both synthetic and real images, in both 2D and 3D, show that our new hybrid method outperforms both the MRF-based and the deformable model-based methods using onlyhomogeneous constraints.

UR - http://www.scopus.com/inward/record.url?scp=34147215974&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=34147215974&partnerID=8YFLogxK

U2 - 10.1007/978-3-540-68020-8_2

DO - 10.1007/978-3-540-68020-8_2

M3 - Chapter

SN - 3540680195

SN - 9783540680192

T3 - Studies in Computational Intelligence

SP - 43

EP - 63

BT - Applied Graph Theory in Computer Vision and Pattern Recognition

A2 - Kandel, Abraham

A2 - Bunke, Horst

A2 - Last, Mark

ER -

Huang R, Pavlovic V, Metaxas D. A graphical model framework for image segmentation. In Kandel A, Bunke H, Last M, editors, Applied Graph Theory in Computer Vision and Pattern Recognition. 2007. p. 43-63. (Studies in Computational Intelligence). https://doi.org/10.1007/978-3-540-68020-8_2