Modulating the Shape and Size of Backprojection Surfaces to Improve Accuracy in Volumetric Stereo

In 3D TV applications, the extraction of 3D representations of dynamic scenes from images plays a central role in the preparation of the presented visual content. This paper focuses on the stereo cue to the extraction of these representations and, in particular, on the recently developed family of volumetric approaches to stereo. Two methods are proposed that improve the accuracy of volumetric stereo approaches, which compare backprojections of image regions to establish stereo correspondences. The proposed methods are based on maximizing the utilization of the available image resolution, as well as, equalizing the sampled image area across pairs of image regions that are compared.