Free-viewpoint scalable multi-view video coding using panoramic mosaic depth maps

In this paper, we propose an efficient scalable multi-view video coding with a novel algorithm to estimate panoramic mosaic depth maps. Multiple view-dependent depth maps are generated by applying a forward depth projection to decoded panoramic mosaic depth maps in the proposed codec. By using the generated depth maps, the proposed scheme not only improves view synthesis prediction performances, but also achieves free-viewpoint scalability and coarse granular SNR scalability. These functionalities will be important to support any kinds of 3D display by a single bitstream and to realize free-viewpoint television. Experiments show that the proposed scheme achieves about 7.4% bitrate reduction relative to the most popular multi-view video coding method and also achieves almost the same efficiency as the conventional scheme which supports neither free-viewpoint scalability nor SNR scalability. Compared to one of the simplest free-viewpoint scalable multi-view coding, up to 16.8% bitrate reduction is achieved by the proposed method.