Learning Edge-Specific Kernel Functions for Pairwise Graph Matching

Michael Donoser, Martin Urschler, Horst Bischof

Research output: Chapter in Book/Report/Conference proceedingConference contributionResearchpeer-review

Abstract

In this paper we consider the pairwise graph matching problem of finding correspondences between two point sets using unary and pairwise potentials, which analyze local descriptor similarity and geometric compatibility. Recently, it was shown that it is possible to learn optimal parameters for the features used in the potentials, which significantly improves results in supervised and unsupervised settings. It was demonstrated that even linear assignments (not considering geometry) with well learned potentials may improve over state-of-the-art quadratic assignment solutions. In this paper we extend this idea by directly learning edge-specific kernels for pairs of nodes. We define the pairwise kernel functions based on a statistical shape model that is learned from labeled training data. Assuming that the setting of graph matching is a priori known, the learned kernel functions allow to significantly improve results in comparison to general graph matching. We further demonstrate the applicability of game theory based evolutionary dynamics as effective and easy to implement approximation of the underlying graph matching optimization problem. Experiments on automatically aligning a set of faces and feature-point based localization of category instances demonstrate the value of the proposed method.
LanguageEnglish
Title of host publicationProceedings of the British Machine Vision Conference
EditorsRichard Bowden, John Collomosse, Krystian Mikolajczyk
PublisherThe British Machine Vision Association
Pages17.1-17.12
ISBN (Print)1-901725-46-4
DOIs
StatusPublished - 2012

Fingerprint

Game theory
Geometry
Experiments

Fields of Expertise

  • Information, Communication & Computing

Cite this

Donoser, M., Urschler, M., & Bischof, H. (2012). Learning Edge-Specific Kernel Functions for Pairwise Graph Matching. In R. Bowden, J. Collomosse, & K. Mikolajczyk (Eds.), Proceedings of the British Machine Vision Conference (pp. 17.1-17.12). The British Machine Vision Association. https://doi.org/10.5244/C.26.17

Learning Edge-Specific Kernel Functions for Pairwise Graph Matching. / Donoser, Michael; Urschler, Martin; Bischof, Horst.

Proceedings of the British Machine Vision Conference. ed. / Richard Bowden; John Collomosse; Krystian Mikolajczyk. The British Machine Vision Association, 2012. p. 17.1-17.12.

Research output: Chapter in Book/Report/Conference proceedingConference contributionResearchpeer-review

Donoser, M, Urschler, M & Bischof, H 2012, Learning Edge-Specific Kernel Functions for Pairwise Graph Matching. in R Bowden, J Collomosse & K Mikolajczyk (eds), Proceedings of the British Machine Vision Conference. The British Machine Vision Association, pp. 17.1-17.12. https://doi.org/10.5244/C.26.17
Donoser M, Urschler M, Bischof H. Learning Edge-Specific Kernel Functions for Pairwise Graph Matching. In Bowden R, Collomosse J, Mikolajczyk K, editors, Proceedings of the British Machine Vision Conference. The British Machine Vision Association. 2012. p. 17.1-17.12 https://doi.org/10.5244/C.26.17
Donoser, Michael ; Urschler, Martin ; Bischof, Horst. / Learning Edge-Specific Kernel Functions for Pairwise Graph Matching. Proceedings of the British Machine Vision Conference. editor / Richard Bowden ; John Collomosse ; Krystian Mikolajczyk. The British Machine Vision Association, 2012. pp. 17.1-17.12
@inproceedings{651b213dbdb64ba4aca398fd8cb763c0,
title = "Learning Edge-Specific Kernel Functions for Pairwise Graph Matching",
abstract = "In this paper we consider the pairwise graph matching problem of finding correspondences between two point sets using unary and pairwise potentials, which analyze local descriptor similarity and geometric compatibility. Recently, it was shown that it is possible to learn optimal parameters for the features used in the potentials, which significantly improves results in supervised and unsupervised settings. It was demonstrated that even linear assignments (not considering geometry) with well learned potentials may improve over state-of-the-art quadratic assignment solutions. In this paper we extend this idea by directly learning edge-specific kernels for pairs of nodes. We define the pairwise kernel functions based on a statistical shape model that is learned from labeled training data. Assuming that the setting of graph matching is a priori known, the learned kernel functions allow to significantly improve results in comparison to general graph matching. We further demonstrate the applicability of game theory based evolutionary dynamics as effective and easy to implement approximation of the underlying graph matching optimization problem. Experiments on automatically aligning a set of faces and feature-point based localization of category instances demonstrate the value of the proposed method.",
author = "Michael Donoser and Martin Urschler and Horst Bischof",
year = "2012",
doi = "10.5244/C.26.17",
language = "English",
isbn = "1-901725-46-4",
pages = "17.1--17.12",
editor = "Richard Bowden and John Collomosse and Krystian Mikolajczyk",
booktitle = "Proceedings of the British Machine Vision Conference",
publisher = "The British Machine Vision Association",
address = "United Kingdom",

}

TY - GEN

T1 - Learning Edge-Specific Kernel Functions for Pairwise Graph Matching

AU - Donoser, Michael

AU - Urschler, Martin

AU - Bischof, Horst

PY - 2012

Y1 - 2012

N2 - In this paper we consider the pairwise graph matching problem of finding correspondences between two point sets using unary and pairwise potentials, which analyze local descriptor similarity and geometric compatibility. Recently, it was shown that it is possible to learn optimal parameters for the features used in the potentials, which significantly improves results in supervised and unsupervised settings. It was demonstrated that even linear assignments (not considering geometry) with well learned potentials may improve over state-of-the-art quadratic assignment solutions. In this paper we extend this idea by directly learning edge-specific kernels for pairs of nodes. We define the pairwise kernel functions based on a statistical shape model that is learned from labeled training data. Assuming that the setting of graph matching is a priori known, the learned kernel functions allow to significantly improve results in comparison to general graph matching. We further demonstrate the applicability of game theory based evolutionary dynamics as effective and easy to implement approximation of the underlying graph matching optimization problem. Experiments on automatically aligning a set of faces and feature-point based localization of category instances demonstrate the value of the proposed method.

AB - In this paper we consider the pairwise graph matching problem of finding correspondences between two point sets using unary and pairwise potentials, which analyze local descriptor similarity and geometric compatibility. Recently, it was shown that it is possible to learn optimal parameters for the features used in the potentials, which significantly improves results in supervised and unsupervised settings. It was demonstrated that even linear assignments (not considering geometry) with well learned potentials may improve over state-of-the-art quadratic assignment solutions. In this paper we extend this idea by directly learning edge-specific kernels for pairs of nodes. We define the pairwise kernel functions based on a statistical shape model that is learned from labeled training data. Assuming that the setting of graph matching is a priori known, the learned kernel functions allow to significantly improve results in comparison to general graph matching. We further demonstrate the applicability of game theory based evolutionary dynamics as effective and easy to implement approximation of the underlying graph matching optimization problem. Experiments on automatically aligning a set of faces and feature-point based localization of category instances demonstrate the value of the proposed method.

U2 - 10.5244/C.26.17

DO - 10.5244/C.26.17

M3 - Conference contribution

SN - 1-901725-46-4

SP - 17.1-17.12

BT - Proceedings of the British Machine Vision Conference

A2 - Bowden, Richard

A2 - Collomosse, John

A2 - Mikolajczyk, Krystian

PB - The British Machine Vision Association

ER -