[Zurück]

M. Nezveda:
"Coarse-to-fine stereo matching for high resolution images";
Poster: Training School on 3D AudioVisual Content Processing and Communications (3D-AVCom 2015), Lisbon, Portugal; 06.07.2015 - 08.07.2015.

Stereo matching denotes the problem of finding correspondences in pairs of images in order to perform 3D reconstruction. However, high resolution images acquired by modern cameras impose difficulties for stereo matching such as computational complexity and matching ambiguities. In this work, we investigate a coarse-to-fine stereo matching framework to resolve the aforementioned difficulties. In this framework, a coarse disparity map guides the disparity estimation in the next level estimation based on an offset map. The offset map determines the disparity search range for finer levels based on the priors of the coarser levels. First experimental results on the Middlebury datasets demonstrate that our coarse-to-fine scheme produces high-quality disparity maps and reduces errors in ambiguous regions when compared to a single-resolution scheme.