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

Citations

1548	F: Conditional Random Fields: Probabilistic models for segmenting and labeling sequence data - Lafferty, McCallum, et al. - 2001
279	Constructing free-energy approximations and generalized belief propagation algorithms - Yedidia, Freeman, et al.
234	Contour and texture analysis for image segmentation - Malik, Belongie, et al. - 2001
189	Combined object categorization and segmentation with an implicit shape model - Leibe, Leonardis, et al. - 2004
165	Minimax entropy principle and its applications to texture modeling - Zhu, Wu, et al. - 1997
142	Efficiently inducing features or conditional random fields - McCallum - 2003
132	Discriminative random fields: A discriminative framework for contextual interaction in classification - Kumar, Herbert - 2003
128	Multiscale conditional random fields for image labeling - He, Zemel, et al.
108	Class-specific, top-down segmentation - Borenstein, Ullman - 2002
103	S.: Combined top-down/bottom-up segmentation - Borenstein, Ullman
95	Conditional random fields for object recognition - Quattoni, Collins, et al. - 2004
90	N.: LOCUS: learning object classes with unsupervised segmentation - Winn, Jojic
53	Kernel conditional random fields: Representation and clique selection - Lafferty, Zhu, et al. - 2004
50	Graph partition by Swendsen-Wang cuts - Barbu, Zhu
40	Segmentation and Boundary Detection Using Multiscale Intensity Measurements - Sharon, Brandt, et al. - 2001
37	Deformable templates - Yuille, Hallinan - 1992
33	Object-specific figure-ground segregation - Yu, Shi - 2003
30	Tree-reweighted belief propagation and approximate ML estimation by pseudo-moment matching - Wainwright, Jaakkola, et al. - 2003
29	Learning and incorporating top-down cues in image segmentation - He, Zemel, et al. - 2006
18	Cue integration in figure/ground labeling - Ren, Fowlkes, et al. - 2005
11	Image parsing: segmentation, detection, and recognition - Tu, Chen, et al. - 2003
6	Fu Jie Huang. Loss functions for discriminative training of energy-based models - LeCun - 2005