The aim of stereo-algorithms is to find corresponding points in two horizontally displaced images. The displacement between these corresponding pixels is called disparity. The disparities vary over the image according to the scene depth, which makes it possible to infer depth maps. When using color images as input for stereo algorithms one would expect higher quality results than when using gray-scale images. However, previous experimental studies found that the use of color images together with brightness-invariant stereo correspondence measures can even have a negative influence on the results. However, these experiments were conducted on a dataset that comprises of images that where captured with the same camera under similar lighting conditions. This raises the question whether these findings can be directly applied to images obtained with different cameras and illumination conditions. Therefore, the aim of this master thesis is to verify the insights of previous works through a new dataset that comprises of images recorded with multiple different cameras. To this end, we first captured stereo image pairs with a variety of different cameras. For the captured stereo pairs we computed reference solutions, in the form of disparity maps, by using a structured light technique. Afterwards, the captured stereo image pairs were used to compute disparity maps with a state-of-the-art stereo matching algorithm. The disparity maps were computed in different color spaces and a variety of correspondence measures were used. Finally, these disparity maps were compared with the reference solutions to obtain error rates. The analysis of the error rates confirms the major insights of previous works: using color in radiometric undistorted image regions only has a minor positive influence on the results. In contrast, radiometric insensitive correspondence measurements can improve the result in these regions. Furthermore, this work confirms that radiometric insensitive correspondence measurements applied on gray-scale images provide better results than if applied on color images.
A. Hasslacher: "Eine Untersuchung zur Wichtigkeit von Farbe in Stereo-Matching"; Supervisor: M. Gelautz, C. Rhemann, M. Bleyer; Fakultät für Informatik der Technischen Universität Wien, 2012.
Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.