On the Video Importance of Multiple Reference Frames

Multiple reference frame prediction (MRFP) is a new function adopted by H.264. This article studies the impact of MRFP to importance and prioritization of video frames. MRFP allows the use of long-term-memory reference frames and multiple-hypotheses signals for motion prediction and compensation, and can thus greatly change the error propagation behaviors of a lost video frame and reorder its importance level. Based on this, we propose a novel video importance scheme called multiple reference video index (MRFI) and present two MRFP-induced effects, i.e., importance weakening of I-frames (I-Weakening) and importance randomization of P-frames (P-Randomization). Comparisons of transmission robustness between MRFI and the existing schemes including EPL and I/P is also presented, and our results show the superiority of MRFI over the existing ones. In particular, the small performance gap between MRFI and EPL in the context of single reference frame is greatly amplified in the context of five reference frames.