## Symmetric stereo matching for occlusion handling (2005)

### Cached

### Download Links

- [research.microsoft.com]
- [research.microsoft.com]
- DBLP

### Other Repositories/Bibliography

Venue: | In CVPR |

Citations: | 111 - 3 self |

### BibTeX

@INPROCEEDINGS{Sun05symmetricstereo,

author = {Jian Sun and Yin Li and Sing Bing and Kang Heung-yeung Shum},

title = {Symmetric stereo matching for occlusion handling},

booktitle = {In CVPR},

year = {2005},

pages = {399--406}

}

### Years of Citing Articles

### OpenURL

### Abstract

In this paper, we propose a symmetric stereo model to handle occlusion in dense two-frame stereo. Our occlusion reasoning is directly based on the visibility constraint that is more general than both ordering and uniqueness constraints used in previous work. The visibility constraint requires occlusion in one image and disparity in the other to be consistent. We embed the visibility constraint within an energy minimization framework, resulting in a symmetric stereo model that treats left and right images equally. An iterative optimization algorithm is used to approximate the minimum of the energy using belief propagation. Our stereo model can also incorporate segmentation as a soft constraint. Experimental results on the Middlebury stereo images show that our algorithm is state-of-the-art. 1

### Citations

1469 | Mean Shift: A Robust Approach toward Feature Space Analysis
- Comaniciu, Meer
- 2002
(Show Context)
Citation Context ...egmentation-based approach does not suffer from this problem because the segment constraint is soft. This advantage is evident in Figure 4. Figure 4(a) shows the mean-shift color segmentation results =-=[5]-=- using default parameters. In Figure 4(b), the gray pixels are reliable pixels selected by robust 3D plane fitting algorithm. These pixels are subsequently used for 3D plane estimation. We ignore regi... |

1384 | Fast Approximate Energy Minimization via Graph Cuts, Proc
- Boykov, Veksler, et al.
- 2001
(Show Context)
Citation Context ...s ∈ {0, 1} is a binary variable. In this paper, we denote � D and � O as disparity and occlusion in a single view, either left or right. Most local stereo algorithms [17] and global stereo algorithms =-=[4, 19]-=- only compute the disparity and occlusion in the reference view (say, the left image). In global algorithm, stereo matching can be formulated in an energy minimization framework: D IR E(DL; I) = Ed(DL... |

1032 | A taxonomy and evaluation of dense two-frame stereo correspondence algorithms
- Scharstein, Szeliski
- 2002
(Show Context)
Citation Context ...ovide reasonable guesses of depths at occluded areas. 1.1 Previous work For a comprehensive discussion on dense two-frame stereo matching, we refer the reader to the survey by Scharstein and Szeliski =-=[17]-=-. In this paper, we review two-frame stereo algorithms that can handle occlusion. Two kinds of hard constraints are typically used: ordering constraint and uniqueness constraint. The ordering constrai... |

340 | Efficient belief propagation for early vision
- Felzenszwalb, Huttenlocher
- 2004
(Show Context)
Citation Context ...T controls the limit of the cost. There are two advantages of using this robust function. For one, it preserves discontinuity. The other advantage is that we can apply a very efficient implementation =-=[8]-=- of belief propagation to minimize the energy involving this smoothness term (described in Section 3).s2.3 Visibility term In order to enforce the visibility consistency constraint, we define the visi... |

266 | Computing visual correspondence with occlusions using graph cuts
- Kolmogorov, Zabih
- 2001
(Show Context)
Citation Context ...lusion is that of cross-checking [7]. In cooperative algorithm [23], the uniqueness constraint is enforced within a 3D array of match values in disparity space using an iterative update algorithm. In =-=[12]-=-, stereo matching is formulated as finding a subset of assignments that may potentially correspond by graph cuts algorithm. A Potts energy to be minimized is defined on the assignments rather than pix... |

265 |
Cooperative computation of stereo disparity
- Marr, Poggio
- 1976
(Show Context)
Citation Context ...g scanlines in both input images. The monotonicity constraint [9] is a variant of the ordering constraint which requires neighboring pixels to be matched. On the other hand, the uniqueness constraint =-=[15]-=- enforces a one-to-one mapping between pixels in two images. 1 Microsoft Research Asia 2 Microsoft Research Beijing, P.R. China Redmond, WA, USA Most approaches [1, 6, 11, 3] that exploit the ordering... |

257 | Multi-camera scene reconstruction via graph cuts
- Kolmogorov, Zabih
- 2002
(Show Context)
Citation Context ...d [14] 1.58 8 1.06 10 8.82 9 0.34 5 0.00 1 3.35 5 1.52 11 2.96 20 2.62 3 0.37 11 5.24 11 Belief prop [19] 1.15 2 0.42 4 6.31 2 0.98 11 0.30 15 4.83 9 1.00 7 0.76 5 9.13 14 0.84 17 5.27 12 MultiCam GC =-=[13]-=- 1.85 11 1.94 16 6.99 6 0.62 9 0.00 1 6.86 13 1.21 9 1.96 11 5.71 7 0.31 8 4.34 10 Region-progress. [22] 1.44 7 0.55 7 8.18 8 0.24 3 0.00 1 2.64 3 0.99 6 1.37 9 6.40 9 1.49 21 17.11 28 GC+occl. [12] 1... |

252 | Stereo matching using belief propagation
- Zheng, Shum
- 2003
(Show Context)
Citation Context ...s ∈ {0, 1} is a binary variable. In this paper, we denote � D and � O as disparity and occlusion in a single view, either left or right. Most local stereo algorithms [17] and global stereo algorithms =-=[4, 19]-=- only compute the disparity and occlusion in the reference view (say, the left image). In global algorithm, stereo matching can be formulated in an energy minimization framework: D IR E(DL; I) = Ed(DL... |

196 | A maximum likelihood stereo algorithm,” Computer vision and image understanding
- Cox, Hingorani, et al.
- 1996
(Show Context)
Citation Context ...er hand, the uniqueness constraint [15] enforces a one-to-one mapping between pixels in two images. 1 Microsoft Research Asia 2 Microsoft Research Beijing, P.R. China Redmond, WA, USA Most approaches =-=[1, 6, 11, 3]-=- that exploit the ordering constraint or uniqueness constraint use dynamic programming. Stereo matching is formulated as finding a minimumcost path in the matrix of all pairwise matching costs between... |

133 | Comparison of Graph Cuts with Belief Propagation for Stereo, using Identical MRF Parameters
- M, Freeman
- 2003
(Show Context)
Citation Context ... for node s and ψst(·) is the compatibility function between node s and t. Belief propagation is an approximate algorithm to minimize the Gibbs energy (7). Recent empirical results on stereo matching =-=[19, 21]-=- show that BP often gives a very good approximation. 3.2 Iterative optimization The optimization process has two steps: 1) estimate occlusion given disparity, and 2) estimate disparity given occlusion... |

128 | Occlusions and binocular stereo
- Geiger, Ladendorf, et al.
- 1995
(Show Context)
Citation Context ... kinds of hard constraints are typically used: ordering constraint and uniqueness constraint. The ordering constraint preserves order along scanlines in both input images. The monotonicity constraint =-=[9]-=- is a variant of the ordering constraint which requires neighboring pixels to be matched. On the other hand, the uniqueness constraint [15] enforces a one-to-one mapping between pixels in two images. ... |

111 | Large occlusion stereo
- Bobick, Intille
- 1999
(Show Context)
Citation Context ...er hand, the uniqueness constraint [15] enforces a one-to-one mapping between pixels in two images. 1 Microsoft Research Asia 2 Microsoft Research Beijing, P.R. China Redmond, WA, USA Most approaches =-=[1, 6, 11, 3]-=- that exploit the ordering constraint or uniqueness constraint use dynamic programming. Stereo matching is formulated as finding a minimumcost path in the matrix of all pairwise matching costs between... |

107 |
A Bayesian approach to binocular stereopsis
- Belhumeur
- 1996
(Show Context)
Citation Context ...er hand, the uniqueness constraint [15] enforces a one-to-one mapping between pixels in two images. 1 Microsoft Research Asia 2 Microsoft Research Beijing, P.R. China Redmond, WA, USA Most approaches =-=[1, 6, 11, 3]-=- that exploit the ordering constraint or uniqueness constraint use dynamic programming. Stereo matching is formulated as finding a minimumcost path in the matrix of all pairwise matching costs between... |

102 |
A global matching framework for stereo computation
- Tao, Sawhney, et al.
(Show Context)
Citation Context ...er is typically 2 or 3 in our experiments. Our results in Figure 5 appear to be very close to the ground truth. 4 Segmentation as Soft Constraint Recently, segmentation-based stereo approaches (e.g., =-=[20, 10, 2, 14, 22]-=-) have demonstrated that the difficulties and ambiguities caused by textureless or occlusion can be handled by using groups of pixels with similar colors. Such approaches assume that there are no larg... |

98 | A Cooperative Algorithm for Stereo Matching and Occlusion Detection
- Zitnick, Kanade
- 1999
(Show Context)
Citation Context ...the path correspond to the left and right occlusions, respectively. The simplest method that uses the uniqueness constraint to detect occlusion is that of cross-checking [7]. In cooperative algorithm =-=[23]-=-, the uniqueness constraint is enforced within a 3D array of match values in disparity space using an iterative update algorithm. In [12], stereo matching is formulated as finding a subset of assignme... |

67 | Segment-based stereo matching using graphcuts
- Hong, G
- 2004
(Show Context)
Citation Context ...er is typically 2 or 3 in our experiments. Our results in Figure 5 appear to be very close to the ground truth. 4 Segmentation as Soft Constraint Recently, segmentation-based stereo approaches (e.g., =-=[20, 10, 2, 14, 22]-=-) have demonstrated that the difficulties and ambiguities caused by textureless or occlusion can be handled by using groups of pixels with similar colors. Such approaches assume that there are no larg... |

67 | Surfaces with occlusions from layered stereo
- Lin, Tomasi
(Show Context)
Citation Context ...er is typically 2 or 3 in our experiments. Our results in Figure 5 appear to be very close to the ground truth. 4 Segmentation as Soft Constraint Recently, segmentation-based stereo approaches (e.g., =-=[20, 10, 2, 14, 22]-=-) have demonstrated that the difficulties and ambiguities caused by textureless or occlusion can be handled by using groups of pixels with similar colors. Such approaches assume that there are no larg... |

66 |
Detecting binocular half-occlusions : empirical comparisons of five approaches
- Egnal, Wildes
- 2002
(Show Context)
Citation Context ...“vertical” discontinuities of the path correspond to the left and right occlusions, respectively. The simplest method that uses the uniqueness constraint to detect occlusion is that of cross-checking =-=[7]-=-. In cooperative algorithm [23], the uniqueness constraint is enforced within a 3D array of match values in disparity space using an iterative update algorithm. In [12], stereo matching is formulated ... |

55 | Gool, “Wide-baseline stereo from multiple views: a probabilistic account - Strecha, Fransens, et al. - 2004 |

33 | A layered stereo algorithm using image segmentation and global visibility constraints
- Bleyer, Gelautz
(Show Context)
Citation Context |

30 | Stereo correspondence with slanted surfaces: critical implications of horizontal slant
- Ogale, Aloimonos
(Show Context)
Citation Context ...arity, the uniqueness constraint is not appropriate for scenes containing horizontally slanted surfaces (Figure 1(a)), which result in correspondence between unequal numbers of pixels. As observed in =-=[16]-=-, a horizontally slanted surface will appear more horizontally stretched in one image compared with the other. Figure 1(b) is the output of the graph cuts [12] algorithm (www.cs.cornell.edu/people/vnk... |

28 | Region-based progressive stereo matching
- Wei, Quan
(Show Context)
Citation Context |

1 |
Occlusions, discontinuities, and epipolar lines
- Ishikawa, Geiger
- 1998
(Show Context)
Citation Context |