View-based 3D video streaming requires the compression and transmission of depth maps along with video sequence. The requirements for representing these depth maps include moderately high compression for fast transmission, preservation of depth discontinuities for high quality video rendering, low complexity decoding, and to be in a form that is suitable for real-time rendering using graphics card. This motivates a mesh-based representation for the depth map. In this paper, we adapt the mesh generation algorithm used in real-time height field rendering for depth representation and produce a regular mesh for each depth map. Then we propose a compression techniques that encodes the triangulated depth map efficiently. The depth map are encoded as a set of single bit enabled flags that mark the split of coarse triangles and depth values at the vertices which are encoded with Huffman code. By coding the vertices in rendering order, we achieve depth map decoding and stripped triangles for OpenGL based rendering in one loop, thus greatly reducing the complexity at the rendering end. Our experiments show great improvement in rendering speed while maintaining high quality for the rendered virtual view.
