Frame-recursive block-based distortion estimation model for multiple reference frames and motion copy concealment in H.264/AVC

Abstract

In this paper, we present an accurate channel distortion estimation model for robust H.264/AVC video codec. The model extends our previously proposed hybrid frame-recursive block-based distortion model by taking into account both multiple reference frames and motion copy concealment techniques which are first introduced in the latest video compression recommendation H.264/AVC. In order to capture the characteristics of multiple reference frames, we introduce a new parameter to relate the dependency of the current frame to more-than-one reference frames and applying a linear combination method to form a statistical model based on previously calculated distortion results. By applying Bernoulli process to approximate the first moment of decoded pixel, we can obtain an estimation formula for both frame copy and motion copy concealments. The simulation results are compared with our proposed channel-error distortion estimation results and the comparisons show that our model yields a fairly high estimation accuracy.