Generalized view synthesis prediction for 3D-AVC

Depth image based rendering technique can be used for 3D video coding in a way that the synthesized view is used as additional prediction signal. In AVC-based 3D video coding (3D-AVC), view synthesis prediction (VSP) technique is utilized for coding non-base views, by adding a virtual synthesized view in the reference picture list. To further improve the efficiency of VSP, in this paper, we propose a generalized view synthesis prediction (GVSP) method which extends the concept of VSP to remove redundant signaling of motion information. With the proposed GVSP, for each macroblock partition, a GVSP mode (with a flag) is introduced to signal whether the VSP is utilized. Experimental results show that, the proposed GVSP method is able to achieve average 0.7% BD rate reduction for 3D-AVC codec with no complexity increase at the decoder. Part of the proposed technique has been adopted by 3D-AVC.