Segment-Based Stereo Matching Using Belief Propagation and a Self-Adapting Dissimilarity Measure

Cooperative Stereo Matching with Color-Based Adaptive Local Support

“…Good results can be achieved e.g. with methods performing an initial color-based segmentation to group pixels belonging to the same object surface [4,5,6]. In this paper we show that a computationally expensive pre-segmentation is not necessary and that the grouping can be coded within pre-calculated weights of local support areas used in a cooperative optimization suitable for later hardware acceleration.…”

“…Surprisingly, grouping neighboring pixels that are assumed to be located on the same object surface by the similarity of their color is relatively new in stereo vision. Recent algorithms use this constraint to obtain local adaptive correlation windows which are better aligned to object borders, resulting in better correlation accuracy at disparity discontinuities [4,5,6,3]. Grouping is achieved either by color segmentation [4,5,6] or by calculating color-dependent correlation weights [3] to control the influence of pixels inside a correlation window on the matching score.…”

“…Recent algorithms use this constraint to obtain local adaptive correlation windows which are better aligned to object borders, resulting in better correlation accuracy at disparity discontinuities [4,5,6,3]. Grouping is achieved either by color segmentation [4,5,6] or by calculating color-dependent correlation weights [3] to control the influence of pixels inside a correlation window on the matching score. Unfortunately, relatively large correlation windows are needed by these algorithms to eliminate ambiguities (in [3] the typical window size is 33x33 pixels, in [6] results are given for 51x51 windows), which results in a higher computational cost making these approaches unsuitable for real-time applications.…”

Recovering consistent video depth maps via bundle optimization

“…By assuming that the neighboring pixels with similar colors have similar or continuous depth values, segmentationbased approach [20,21,9] can improve the depth estimation especially for large textureless regions. These methods typically model each segment as a 3D plane and estimate the plane parameters by matching small patches between neighboring viewpoints [21,9], or using a robust fitting algorithm [20].…”

Dynamically adaptive real-time disparity estimation hardware using iterative refinement

“…These are ranked with respect to their performance in the evaluation of Middlebury benchmarks [11]. Although top-performer algorithms provide impressive visual and quantitative results [12][13][14], their implementations in real-time High Resolution (HR) stereo video are challenging due to their complex multi-step refinement processes or their global processing requirements that demand huge memory size and bandwidth. For example, the AD-Census algorithm [12], currently the top published performer, provides successful results that are very close to the ground truths.…”