Fast disparity and motion estimation based on correlations for multiview video coding

Recently, multiview video coding has attracted great attention from industries and research institutes. However, the heavy computational complexity limits its practical applications. In this paper, a fast disparity and motion estimation for multiview video coding is presented, based on the correlations between the neighboring cameras and between the motion and the disparity. In the proposed approach, first, a search region estimation is proposed to reduce the disparity estimation complexity according to that the camera set is usually fixed and therefore the disparity between the two neighboring views can be limited to an estimable range. Second, a motion vector derivation is given based on the geometric relationship between the motion and the disparity. In addition, an early termination scheme is provided to further reduce the number of reference frames. The experimental results show that roughly 50% time saving for disparity and motion computing can be reached when compared to the anchor in multiview video coding test model JSVM only with negligible coding efficiency loss.

I. Introduction

Multiview video is a group of video sequences captured by a set of cameras on different positions at the same time instance and from the same scene. It is much favored in the services such as free-viewpoint video (FVV), free-viewpoint television (FTV), and 3DTV [1], which expand the user's sensation far beyond what is offered by the traditional video. Multiview video coding (MVC) has attracted great attention from industries and research institutes. Recently, MPEG 3-D audio and video (3DAV) ad hoc group has been investigating the needs for standardization in 3DTV and FTV. The Joint Video Team (JVT) is also working on a MVC extension of H.264/A VC to meet the industry demand.