Efficient Inter-View Motion Vector Prediction in Multi-View HEVC

Abstract

When multi-view video sequences are encoded, the efficient motion vector prediction (MVP) is one of the most important techniques to increase the coding performance. The predictive MV should be highly correlated with the actual MV of the current block to be encoded. In this paper, we explain the inefficiency of the MVP algorithm in the multi-view high efficiency video codec (MV-HEVC) standard. In order to solve the inefficiency problem, we propose an enhanced advanced motion vector prediction algorithm in which the geometry interrelation between two neighboring views is derived based on epipolar geometry, similarity transform, and affine transform, and then predicted motion vectors (PMVs) for efficient MV coding are generated using obtained geometry relation. The experiment results show that the proposed scheme outperforms conventional MV-HEVC with a 1.32%–1.19% coding gain because it provides more efficient PMVs than conventional algorithms.