Stereo Matching with Color-Weighted Correlation, Hierachical Belief Propagation and Occlusion Handling

By Qingxiong Yang, Liang Wang, Ruigang Yang, Henrik Stewénius, David Nistér. Computer Vision and Pattern Recognition, 2006

678 citations recorded by the source corpus.

In this paper, we formulate an algorithm for the stereo matching problem with careful handling of disparity, discontinuity and occlusion. The algorithm works with a global matching stereo model based on an energy- minimization framework. The global energy contains two terms, the data term and the smoothness term. The data term is first approximated by a color-weighted correlation, then refined in occluded and low-texture areas in a repeated application of a hierarchical loopy belief propagation algorithm. The experimental results are evaluated on the Middlebury data set, showing that our algorithm is the top performer.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/7370a13b1ad731a35725e247976e6d7415658a46

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

PatchMatch Stereo - Stereo Matching with Slanted Support Windows (2011, background): “Window-based matching costs allow global algorithms to inherit the ability to precisely capture depth discontinuities from adaptive support weight approaches [15].”
Global stereo reconstruction under second order smoothness priors (2008, methodology): “While their two approaches differ somewhat, both enforce the constraint that segmented regions of the image be planar, a trait common to the sequence of algorithms that succeeded Tao's [4] [15] [17] [35], which have shown excellent results on the " Middlebury " test set [27].”
Fusion of time-of-flight depth and stereo for high accuracy depth maps (2008, methodology): “In our implementation the pixelwise matching cost are obtained using a similar manner as was recently proposed by [24].”
Spatial-Depth Super Resolution for Range Images (2007, methodology): “(a) Depth maps generated with the DoubleBP algorithm [19] reported on the Middlebury website.”
Ieee Transactions on Pattern Analysis and Machine Intelligence 1 Stereo Processing by Semi-global Matching and Mutual Information (methodology): “Almost all of the currently top-ranked algorithms [2], [3], [7], [9], [11]–[15] on the Tsukuba, Venus, Teddy and Cones data set [18] optimize a global energy function.”
Particle Belief Propagation (2009, methodology): “Belief propagation was applied to the stereo vision problem in Sun et al. (2003), and since then a number of more sophisticated models have also used BP for inference (e.g., Sun et al., 2005; Klaus et al., 2006; Yang et al., 2006).”
DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch (2019, background): “Classic stereo approaches typically start by computing robust feature representations [38, 36, 31, 20, 41].”
Pixelwise View Selection for Unstructured Multi-View Stereo (2016, background): “MVS leverages multiple views to overcome the inherent occlusion problems of two-view approaches [35–37].”
High-Accuracy Stereo Matching Based on Adaptive Ground Control Points (2015, positive): “Recently better results were achieved by combining new constraint hypothesis such as color segmentation [11], [12] matting [13], [14] and weighed support window [15].”
Real-time O(1) bilateral filtering (2009, background): “Originally introduced by Tomasi and Manduchi [2], bilateral filters are edge preserving operators that have found widespread use in many computer vision and graphics tasks like denoising [3, 4, 5, 6, 7], texture editing and relighting [8], tone management [9, 10], demosaicking [11], stylization [12]”
Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation (2022, methodology): “Global methods treat stereo matching as an optimization problem, where an explicit cost function is formulated and optimized by belief propagation [20,42,52] or graph cut [4] algorithms.”
A hand pose tracking benchmark from stereo matching (2017, methodology): “There are a lot of stereo matching methods [11–14].”
Consistent Depth Maps Recovery from a Video Sequence (2009, methodology): “By assuming that the neighboring pixels with similar colors have similar depth values, segmentation-based approaches [42], [8], [ 47 ], [21], [40] were proposed to improve the depth estimation for large textureless regions.”
Real-time Global Stereo Matching Using Hierarchical Belief Propagation (2006, methodology): “One thing worth noticing is that most of these methods, such as [ 1 , 11, 6], integrate multiple low-level visual cues (e.g., segmentation, edges, visibility testing) as either soft or hard constraints to improve stereo matching while our approach works under a basic and 6”
Segment-Based Stereo Matching Using Belief Propagation and a Self-Adapting Dissimilarity Measure (2006, methodology): “321 Double-BP [12] 2.”