An efficient hole filling for depth image based rendering

An efficient hole filling algorithm is presented providing texture compilation for invisible pixels during depth image based rendering (DIBR). Some pixels cannot be assigned to proper texture values due to occluding foreground object; thus, compilation is required to obtain visually pleasing virtual views from a set of reference views. The proposed algorithm exploits reliable texture transition from visible pixels to the holes among horizontal and vertical directions. Transition is performed by successive weighted summation (SWS) based on color and depth similarities between the neighboring pixels, producing adaptively weighted and connected aggregation. Depending on local depth variation, assignment of proper texture to the hole pixels is achieved by combining horizontal and vertical information. According to the experimental results, proposed approach yields visually pleasing compilation of occluded regions with drastically decreased number of operations per pixel.