#### DMCA

## Symmetric sub-pixel stereo matching (2002)

### Cached

### Download Links

- [www.middlebury.edu]
- [community.middlebury.edu]
- [www.cs.middlebury.edu]
- [www.cs.middlebury.edu]
- [research.microsoft.com]
- [research.microsoft.com]
- DBLP

### Other Repositories/Bibliography

Citations: | 30 - 4 self |

### Citations

2863 | An iterative image registration technique with an application to stereo vision.
- Lucas, Kanade
- 1981
(Show Context)
Citation Context ...sparity map. Alternately, sub-pixel estimates could be recomputed at the end around each winning disparity using one of the techniques described in [17], e.g., using a Lucas-Kanade gradient-based fit =-=[10]-=- to nearby pixels at the same disparity. 4 Experimental evaluation of matching costs Since there are so many alternatives possible for computing the DSI, how do we choose among them? From theoretical ... |

2100 | Fast approximate energy minimization via graph cuts.
- Boykov, Veksler, et al.
- 2001
(Show Context)
Citation Context ...ve seen a dramatic improvement in the quality of dense stereo matching algorithms [14]. A lot of this improvement can be attributed to better optimization algorithms and better smoothness constraints =-=[6, 4, 16]-=-. However, a remarkable amount of the improvement has also come from better matching metrics at the input [3]. In fact, Birchfield and Tomasi’s sampling-insensitive dissimilarity measure is used by a ... |

1533 | A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
- Scharstein, Szeliski
- 2002
(Show Context)
Citation Context ...lidate our approach with experimental results on stereo images with ground truth. 1 Introduction The last few years have seen a dramatic improvement in the quality of dense stereo matching algorithms =-=[14]-=-. A lot of this improvement can be attributed to better optimization algorithms and better smoothness constraints [6, 4, 16]. However, a remarkable amount of the improvement has also come from better ... |

526 |
A computational framework an an algorithm for the measurement of visual motion.
- Anandan
- 1989
(Show Context)
Citation Context ...g such interval analysis is to directly compute the squared differences, and to then fit a parabola to the resulting sampled DSI. This is a classic approach to obtaining sub-pixel disparity estimates =-=[17, 1, 11]-=-, although applying it directly to integer-valued displacements (disparities) can lead to severe biases [15]. When the DSI has been adequately sampled, however, this is a very useful alternative for e... |

337 | A stereo matching algorithm with an adaptive window: Theory and experiment
- Kanade, Okutomi
- 1994
(Show Context)
Citation Context ...ered, which is discussed next. 5.3 Propagating matches To propagate certain matches into ambiguous matching regions, we propose an iterative algorithm similar in spirit to adaptive-windows techniques =-=[8, 13, 20]-=-. Our algorithm interleaves the selection of certain matches with further aggregation of the DSI, using successively larger windows. After each aggregation step, new certain matches are selected among... |

257 | Kalman filterbased algorithms for estimating depth from image sequences.
- Matthies, Kanade, et al.
- 1989
(Show Context)
Citation Context ...g better than just pixel-sampled intensity differences is not a new idea. For example, Matthies et al. interpolated scanlines by a factor of 4 using a cubic interpolant before computing the SSD score =-=[11]-=-. Tian and Huhns wrote an even earlier survey paper comparing various algorithms for sub-pixel registration [17]. In fact, some stereo and motion algorithms have always evaluated displacements on a ha... |

206 | A pixel dissimilarity measure that is insensitive to image sampling.
- Birchfield, Tomasi
- 1998
(Show Context)
Citation Context ...n be attributed to better optimization algorithms and better smoothness constraints [6, 4, 16]. However, a remarkable amount of the improvement has also come from better matching metrics at the input =-=[3]-=-. In fact, Birchfield and Tomasi’s sampling-insensitive dissimilarity measure is used by a number of today’s best performing algorithms [6, 4]. Using something better than just pixel-sampled intensity... |

142 | Large occlusion stereo.
- Bobick, Intille
- 1999
(Show Context)
Citation Context ...ed this fact explicitly (P. Anandan, personal communication). The set of initial matching costs that are fed into a stereo matcher’s optimization stage is often called the disparity space image (DSI) =-=[19, 5]-=-. However, while the concept of stereo matching as finding an optimal surface through this space has been around for a while [19, 2, 5], relatively little attention has been paid to the proper samplin... |

137 |
A parallel stereo algorithm that produces dense depth maps and preserves image features’,
- Fua
- 1993
(Show Context)
Citation Context ...certain, i.e., correct with high probability. First, there should be a clear minimum among all candidate cost values. Second, the minima for both left-to-right and right-to-left matching should agree =-=[7, 5]-=-, which can be checked by examining the vertical and diagonal columns in disparity space (Figure 5b). While the second test is easy to implement, the definition of a clear minimum is less straightforw... |

135 | Multiway cut for stereo and motion with slanted surfaces.
- Birchfield, Tomasi
- 1999
(Show Context)
Citation Context ...ve seen a dramatic improvement in the quality of dense stereo matching algorithms [14]. A lot of this improvement can be attributed to better optimization algorithms and better smoothness constraints =-=[6, 4, 16]-=-. However, a remarkable amount of the improvement has also come from better matching metrics at the input [3]. In fact, Birchfield and Tomasi’s sampling-insensitive dissimilarity measure is used by a ... |

125 | Stereo matching with nonlinear diffusion.
- Scharstein, Szeliski
- 1998
(Show Context)
Citation Context ...ered, which is discussed next. 5.3 Propagating matches To propagate certain matches into ambiguous matching regions, we propose an iterative algorithm similar in spirit to adaptive-windows techniques =-=[8, 13, 20]-=-. Our algorithm interleaves the selection of certain matches with further aggregation of the DSI, using successively larger windows. After each aggregation step, new certain matches are selected among... |

124 |
A global matching framework for stereo computation.
- Tao, Sawhney, et al.
- 2001
(Show Context)
Citation Context ...ve seen a dramatic improvement in the quality of dense stereo matching algorithms [14]. A lot of this improvement can be attributed to better optimization algorithms and better smoothness constraints =-=[6, 4, 16]-=-. However, a remarkable amount of the improvement has also come from better matching metrics at the input [3]. In fact, Birchfield and Tomasi’s sampling-insensitive dissimilarity measure is used by a ... |

121 |
A Bayesian approach to binocular stereopsis.
- Belhumeur
- 1996
(Show Context)
Citation Context ... optimization stage is often called the disparity space image (DSI) [19, 5]. However, while the concept of stereo matching as finding an optimal surface through this space has been around for a while =-=[19, 2, 5]-=-, relatively little attention has been paid to the proper sampling and treatment of the DSI. In this paper, we take a more careful look at the structure of the DSI, including its frequency characteris... |

93 | Computing rectifying homographies for stereo vision.
- Loop, Zhang
- 1999
(Show Context)
Citation Context ...a disparity map dL(x, y) such that the two images match as closely as possible IL(x, y) ≈ IR(x − dL(x, y),y). (1) (We assume that the images have been rectified to have a horizontal epipolar geometry =-=[12, 9]-=-. We will also want to impose some smoothness or other prior constraints on the function dL(x, y).) Define the 3D signed difference image (SDI) as the intensity (or color) difference between the shift... |

73 |
Algorithms for subpixel registration.
- Tian, Huhns
- 1986
(Show Context)
Citation Context ...ted scanlines by a factor of 4 using a cubic interpolant before computing the SSD score [11]. Tian and Huhns wrote an even earlier survey paper comparing various algorithms for sub-pixel registration =-=[17]-=-. In fact, some stereo and motion algorithms have always evaluated displacements on a half-pixel grid, but never mentioned this fact explicitly (P. Anandan, personal communication). The set of initial... |

66 |
A multiple baseline stereo
- Okutomi, Kanade
- 1993
(Show Context)
Citation Context ...a disparity map dL(x, y) such that the two images match as closely as possible IL(x, y) ≈ IR(x − dL(x, y),y). (1) (We assume that the images have been rectified to have a horizontal epipolar geometry =-=[12, 9]-=-. We will also want to impose some smoothness or other prior constraints on the function dL(x, y).) Define the 3D signed difference image (SDI) as the intensity (or color) difference between the shift... |

63 | Statistical calibration of ccd imaging process
- Tsin, Ramesh, et al.
- 2001
(Show Context)
Citation Context ... h](xi,yi)+nL(xi,yi) (4) ÎR(xi,yi) =[IR ∗ h](xi,yi)+nR(xi,yi), (5) 526swhere h(x, y) is the combined point-spread-function of the optics and sampling sensor (e.g., it incorporates the CCD fill factor =-=[18]-=-), and nL is the (integrated) imaging noise. Given that we can only evaluate the DSI at the integral grid positions (xi,yi), we have to ask whether this sampling of the DSI adequate, or whether there ... |

26 |
Local, global, and multilevel stereo matching.
- Yang, Yuille, et al.
- 1993
(Show Context)
Citation Context ...ed this fact explicitly (P. Anandan, personal communication). The set of initial matching costs that are fed into a stereo matcher’s optimization stage is often called the disparity space image (DSI) =-=[19, 5]-=-. However, while the concept of stereo matching as finding an optimal surface through this space has been around for a while [19, 2, 5], relatively little attention has been paid to the proper samplin... |

25 | A progressive scheme for stereo matching
- Zhang, Shan
(Show Context)
Citation Context ...gregates the DSI with increasingly larger windows to disambiguate matches in untextured areas. (A related technique that starts with high-confidence corner matches has been proposed by Zhang and Yang =-=[20]-=-.) For easier reasoning about visibility, we collapse the DSI to integer sampling, as discussed in Section 3. 5.1 Selecting certain matches There are two basic tests to determine whether a match is ce... |

21 |
Precise sub-pixel estimation on area-based matching.
- Shimizu, Okutomi
- 2001
(Show Context)
Citation Context ...or near true matches as a better reconstructed DSI. Also, the resulting minima will be sensitive to the original placement of samples, i.e., a significant bias towards integral disparities will exist =-=[15]-=-. For example, if the original signal is a fairly high-frequency chirp (Figure 2a), then applying a piecewise linear interpolant will fail to correctly match the signal with a fractionally shifted ver... |