Low complexity disparity estimation for immersive 3D video transmission

Abstract

Bandwidth-limited channels demand the transmission of the per-pixel depth maps with the texture data to provide immersive 3D video services that allow arbitrary 3D viewpoint reconstruction. This auxiliary depth data offers geometric information, which together with the multi-view and epipolar geometries, can be exploited during 3D video coding to calculate geometric positions for the search areas of disparity estimation. These positions represent a more accurate estimate match to compensate the current macro-block from than those provided by the median adopted by the H.264/MVC standard. The result is smaller search areas that reduce the encoder's computational requirement. In this work, we exploit this fact together with the largest depth variation within the macro-block to encode, to calculate and adaptively adjust these areas along the epipolar lines. The proposed solution achieves a speedup gain of up-to 32 times over the original disparity estimation, with negligible influence on the rate-distortion performance of 3D video coding. This highly reduces the computational cost of the H.264/MVC encoder and eases its need to be implemented on highly expensive systems that are otherwise necessary to meets the stringent latency requirement of broadcast transmissions. Moreover, it provides similar coding efficiencies required for such scenarios.