A Novel Confidence Measure for Disparity Maps by Pixel-Wise Cost Function Analysis

Disparity estimation algorithms mostly lack information about the reliability of the disparities. Therefore, errors in initial disparity maps are propagated in consecutive processing steps. This is in particularly problematic for difficult scene elements, e.g., periodic structures. Consequently, we introduce a simple, yet novel confidence measure that filters out wrongly computed disparities, resulting in improved final disparity maps. To demonstrate the benefit of this approach, we compare our method with existing state-of-the-art confidence measures and show that we improve the ability to detect false disparities by 54.2%.