Learning to combine bottom-up and top-down segmentation

DMCA

Learning to combine bottom-up and top-down segmentation

Cached

Download Links

Other Repositories/Bibliography

DBLP

by Anat Levin

Venue:	in: European Conference on Computer Vision
Citations:	131 - 0 self

BibTeX

@INPROCEEDINGS{Levin_learningto,
    author = {Anat Levin},
    title = {Learning to combine bottom-up and top-down segmentation},
    booktitle = {in: European Conference on Computer Vision},
    year = {},
    pages = {581--594}
}

OpenURL

Abstract

Abstract. Bottom-up segmentation based only on low-level cues is a notoriously difficult problem. This difficulty has lead to recent top-down segmentation algorithms that are based on class-specific image information. Despite the success of top-down algorithms, they often give coarse segmentations that can be significantly refined using low-level cues. This raises the question of how to combine both top-down and bottom-up cues in a principled manner. In this paper we approach this problem using supervised learning. Given a training set of ground truth segmentations we train a fragment-based segmentation algorithm which takes into account both bottom-up and top-down cues simultaneously, in contrast to most existing algorithms which train top-down and bottom-up modules separately. We formulate the problem in the framework of Conditional Random Fields (CRF) and derive a feature induction algorithm for CRF, which allows us to efficiently search over thousands of candidate fragments. Whereas pure top-down algorithms often require hundreds of fragments, our simultaneous learning procedure yields algorithms with a handful of fragments that are combined with low-level cues to efficiently compute high quality segmentations. 1

Keyphrases

top-down segmentation low-level cue training set bottom-up segmentation conditional random field coarse segmentation top-down algorithm simultaneous learning procedure yield bottom-up cue fragment-based segmentation algorithm bottom-up module principled manner top-down cue class-specific image information high quality segmentation feature induction algorithm ground truth segmentation whereas pure top-down algorithm recent top-down segmentation difficult problem candidate fragment

DMCA