A fast projection method for connectivity constraints in image segmentation

Jan Stühmer, Daniel Cremers

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

2 Scopus citations

Abstract

We propose to solve an image segmentation problem with connectivity constraints via projection onto the constraint set. The constraints form a convex set and the convex image segmentation problem with a total variation regularizer can be solved to global optimality in a primal-dual framework. Efficiency is achieved by directly computing the update of the primal variable via a projection onto the constraint set, which results in a special quadratic programming problem similar to the problems studied as isotonic regression methods in statistics, which can be solved with O(n log n) complexity. We show that especially for segmentation problems with long range connections this method is by orders of magnitudes more efficient, both in iteration number and runtime, than solving the dual of the constrained optimization problem. Experiments validate the usefulness of connectivity constraints for segmenting thin structures such as veins and arteries in medical image analysis.

Original languageEnglish
Title of host publicationEnergy Minimization Methods in Computer Vision and Pattern Recognition - 10th International Conference,EMMCVPR 2015, Proceedings
EditorsXue-Cheng Tai, Egil Bae, Tony F. Chan, Marius Lysaker
PublisherSpringer Verlag
Pages183-196
Number of pages14
ISBN (Electronic)9783319146119
DOIs
StatePublished - 2015
Event10th International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition, EMMCVPR 2015 - Hong Kong, China
Duration: 13 Jan 201516 Jan 2015

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume8932
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference10th International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition, EMMCVPR 2015
Country/TerritoryChina
CityHong Kong
Period13/01/1516/01/15

Fingerprint

Dive into the research topics of 'A fast projection method for connectivity constraints in image segmentation'. Together they form a unique fingerprint.

Cite this