A Fast Multi-reference Frame Selection Algorithm for Multiview Video Coding

Multiview video coding (MVC) plays an important role in three-dimensional video applications. Joint Video Team developed a joint multiview video model (JMVM) with multi-reference frame technology. Motion and disparity estimation are employed in multi-reference frame technology to provide better rate distortion performance. However, the process of searching blocks with variable sizes for motion and disparity estimation in multi-reference frames significantly increases the computational complexity. After analyzing the statistical features of multi-reference frames in MVC, this paper proposes a fast multi-reference frame selection algorithm based on a dynamic threshold technology correlated with the types of frames. All views of prediction structure in JMVM are categorized into three types, that is, the basic view without inter-view reference relation, the first layer views that only refer anchor frame from upper view, the second layer views that need to refer the basic view and the first layer views. Furthermore, all frames in MVC prediction structure are divided into three categories, the anchor frame without any reference, non-anchor frame in the base view and the first layer view, and the non-anchor frame in other views. Dynamic threshold technique is given to terminate the process of searching multi-reference frames early. The proposed algorithm also reduces the useless candidate reference frames by filtering out those reference frames that are less likely to contain the best matched results and achieves significant speed-up. Experimental results show that the proposed algorithm can achieve 50.13% ~ 72.19% reduction of encoding time in comparison with JMVM7.0, while the proposed algorithm hardly influences the rate distortion performance of MVC.