A novel adaptive prediction structure for multiview video coding

The prediction structure of multi-view video coding (MVC) is expected to support temporal random access, high compression efficiency and low computational complexity. Since the characteristic of multi-view video (MVV) sequences change from sequence to sequence, the fixed prediction structure of Joint Multi-view Video Coding (JMVC) is difficult to cope with various characteristics of multi-view video sequences. To gain better coding performance, an adaptive prediction structure for MVC (APS_MVC) based on temporal and inter-view correlation is proposed in this paper. MVV sequences are partitioned into three categories according to their temporal and inter-view correlation. Consequently, three different MVC prediction structures are developed. The best fit prediction structure is automatically selected according to the temporal and inter-view correlation of the encoded sequence. Experimental results have shown that the proposed APS_MVC outperforms Scalable Prediction Structure Scheme for MVC (SPS_MVC) on reducing computational complexity, improving random access ability, and decreasing decoding picture buffer (DPB) size by about 41%, 30%, and 35% on average, respectively.