Multiview Video Coding Based on Global Motion Model

Xun Guo, Qingming Huang

Abstract: In this paper, we present a novel scheme for coding multiview video sequence based on global motion prediction between adjacent views. For that, the left-most view is compressed as reference sequence using standard block-based motion compensated prediction coding. And its right view is compressed with global motion prediction from the left view images. In the prediction, an eight-parameter global motion model is used to compute the global motion information between left and right-view images. Then the motion vectors of the right-view image are predicted from the left-view image based on the global information. To further reduce the coding complexity and improve coding efficiency, macroblock modes of current image are also predicted from the left-view image. H.264 coding scheme is employed as the baseline, in which Rate-Distortion Optimization is used to select the best coding mode. Experimental results show that, compared to coding multiview video sequence independently, the proposed scheme can save the bitrate up to 15%.