Direct virtual viewpoint synthesis from multiple viewpoints

This paper presents a novel approach for synthesizing intermediate or virtual viewpoints (VVs) of a 3D scene based on information from a number of known reference viewpoints (RVs). The proposed approach directly estimates the pixel value (and corresponding depth) for each pixel in the VV. This is contrast to the more traditional 2 stage approach of firstly building a full 3D or 2.5D model for the scene and then synthesising the desired VV. The potential advantage of this approach is that it works directly with the target virtual view and is hopefully less susceptible to the propagation of errors from the depth estimation stage to the interpolation stage.