Statistical Early Termination Model for Fast Mode Decision and Reference Frame Selection in Multiview Video Coding

Multiview Video Coding (MVC) adopts exhaustive variable size mode decision and multiple reference frame selection to significantly improve high compression efficiency at each macroblock. However, these two technologies increase the computational complexity of MVC encoders tremendously. In this paper, we propose an efficient Statistical DIRECT Mode Early Termination (SDMET) model which estimates the rate distortion degradation, false acceptance rate and false reject rate of early DIRECT mode decision. It can adaptively adjust the rate distortion cost threshold not only according to the quantization parameter, but also the video content and motion properties. Experimental results show that SDMET can reduce 42.40% to 65.60% computation complexity for fast mode decision. When it is jointly optimized with fast multi-reference frame selection, the proposed overall algorithm can achieve 79.57% to 89.21% computational complexity reduction with unnoticeable rate distortion degradation. Additionally, the proposed SDMET and the overall fast mode decision algorithm can be applied to both temporal views and inter-view views in MVC.