Robust Multiple Hypothesis Motion Compensated Prediction within the H.264/AVC standard

Abstract

Multiple Hypothesis Motion Compensated Prediction (MHMCP) has been extensively studied and is known to be a very effective method for enhancing the error resilience of motion compensated prediction with little impact on coding efficiency for compressing moving picture sequences. MHMCP with 2 hypotheses (2HMCP) may be used within the H.264/AVC standard by implementing it under the generalized B pictures framework. We explore good implementations of 2HMCP and propose a novel method to further enhance the error resilience by restricting the search areas in each reference frame such that the hypotheses selected from each are not too correlated. We find that there is significant rate-distortion gain, especially for fast moving sequences.