Stability-based Tree for Disparity Refinement

In this paper, a disparity refinement method with stability-based tree for stereo matching is proposed. In stereo matching, disparity information can be extracted by examining the relative positions of objects in the two panels. We focus on refinement and let this step play an effective role in improving the quality of the disparity map. We observe the features of support regions which produce disparity errors and summarize the reason as unstableness. By developing stability-based tree to evaluate and reconstruct support regions for cost aggregation, the proposed method achieves effective performance in removing outliers. Extensive experiments on both laboratory and re-al-world road datasets demonstrates that the proposed method outperforms existing algorithms in removing large error parts as well as smoothing fractions. It reduces more than 70% aggregation time compared with traditional tree method without loss of accuracy.