Stereoscopic 3D conversion from 2D video in H.264/AVC compressed domain

In this paper, we proposed a depth map generation algorithm for the application in 2D to 3D video conversion. The proposed algorithm employs two main processes to generate a stereoscopic video in H.264/AVC compressed domain. First, the processing of the motion vector is performed to improve the robustness of the motion information, which are directly extracted from the compressed bitstreams. And this process composes three parts: motion vector field initialization, zero motion vector decision and spatial motion vector filtering. Second, the block level depth map is estimated from the filtered motion vectors of all blocks, and it is up-sampled by a joint bilateral filter to get the final depth map with the same resolution as the original video. The rendered anaglyph frames and experiments showed that the proposed algorithm works well for sequences with static and dynamic background.