Stereo Matching with Reliable Disparity Propagation

Abstract

In this paper, we propose a novel propagationbased stereo matching algorithm. Starting from an initial disparity map, our algorithm selects highly reliable pixels and propagates their disparities along the scanline to produce dense disparity results. The key idea is to construct a line segment region for each pixel with local color and connectivity constraints. The pixelwise line segments are efficiently used to compute initial disparities, select reliable pixels and determine proper propagation regions. Streaking artifacts are effectively removed in a refinement process. Experimental results demonstrate the performance of the proposed method: it ranks 5th in the Middlebury benchmark, and the results can be computed within a few seconds.

A hybrid genetic approach for stereo matching

“…Our scheme also enables to "boost" a weak local optimization scheme via a genetic algorithm. Specifically, it could replace scan-line optimization modules in state-of-the-art algo- Figure 7: Top row shows the Venus stereo pair and its ground truth; bottom row shows (from left to right) the resulting disparity maps of our approach and the original scan-line algorithm) Figure 8: Top row shows the Teddy stereo pair and its ground truth; bottom row shows (from left to right) the resulting disparity maps of our approach and the original scan-line algorithm rithms (that achieve higher accuracy than that of our hybrid algorithm) [15,13] to further enhance their performance.…”

Learning to Detect Ground Control Points for Improving the Accuracy of Stereo Matching

“…Wang and Yang [25] pick GCPs by running three different Winner-Take-All (WTA) stereo algorithms and require that the disparities be consistent among all the matchers in each image, as well as leftright consistent. Sun et al [24] used LRC and the ratio of the best to the second best matching cost in a disparity propagation framework. Our approach integrates numerous criteria in a principled way via supervised learning and learns how to make decisions based on labeled data rather than intuition.…”

A Convex Solution to Disparity Estimation from Light Fields via the Primal-Dual Method

“…However, these errors can only influence a small part of the image and are therefore sparse in a canonical basis [2,33]. An alternative to explicit occlusion modeling is to match only reliable pixels and fill the unmatched correspondences via regularization [18,27]. However, as explained in [28], these methods are prone to artifacts.…”