ABSTRACT
Acquired images and video are nowadays often transmitted along with a depth map, or range map, displaying information on the distance of each framed object from the camera. Such information enables various advanced video-based functionalities, ranging from free view point television and 3D video, to advanced video surveillance or computer vision, to cite a few. This paper addresses the problem of upsampling a low-resolution depth map to reconstruct the unknown high resolution depth map. Depth maps acquisition and compression are expected to be key technologies for future innovative video services (1). The problem of depth map upsampling emerges primarily from a technological limit in the range acquisition devices; in fact, due to the high cost of the range map acquisition technology, accurate and robust techniques for improving the resolution of measured depth maps are required. While high definition video and photo cameras are commonly available on the market (e.g. 1920× 1240 or larger), the commercial depth acquisition devices usually provide lower resolution (e.g. 176× 144) maps. Moreover, in order to reduce the bandwidth occupancy of 3D or multi-view video systems, depth maps are usually downsampled before compression, so that a suitable upsampling stage is needed at receiver side.
