Depth-Image Representation Employing Meshes for Intermediate-View Rendering and Coding

A common way to represent 3-D images is to send the texture-image together with a corresponding depth-image. This paper presents a depth-image representation employing triangular meshes to benefit both in obtaining higher compression factors and directly enabling fast intermediate-view rendering on commodity graphics hardware. Previous mesh-based representations did not consider the fact that assigning wrong depth values along object boundaries results in well-observable artifacts in the intermediate-view rendering. We solve this problem by assigning two depth values per node to better model sharp depth discontinuities. Furthermore, we ensure that mesh-edges are placed along object borders, preventing that triangles span background as well as objects at the same time. The results show good intermediate-view rendering quality, object/background separation, and good rate-distortion performance up to about 40 dB PSNR at a bit rate of 0.05 bits/pixel.