Distributed residual coding for multi-view video with joint motion vector projection and 3-D warping

Conventional multi-view video coding (MVC) requires high encoding complexity and high rate data transmission from cameras to a joint encoder. Distributed coding theories guarantee that low complexity and independent encoders can achieve similar coding efficiency as traditional methods. However, there is still a big performance gap between current distributed multi-view video coding (DMVC) schemes and conventional schemes. In this paper, we propose a novel distributed residual coding framework for multi-view video with joint motion vector projection (MVP) and 3-D warping. At the encoder, by means of the improved block classification, distributed residual coding is performed effectively to help reduce temporal redundancy. At the decoder, a joint motion vector projection and 3-D warping method is proposed to generate the candidates of side information and two binary fusion masks. Finally, we employ a novel fusion criterion with relation to the reliability of side information acquisition measures. Experimental results show that the proposed DMVC framework has achieved both side information quality and rate distortion performance improvement.