Abstract
This paper presents a novel error concealment algorithm for H.264/AVC based on a regression model, which is constructed according to the spatial relationship between block locations and their motion activities. With the proposed algorithm, a corrupted macroblock is partitioned into subblocks and the motion vector of each sub-block is predicted through the regression model with the help of the neighbor motion vectors. The experimental results show that the proposed algorithm can achieve significant Peak Signal Noise Ratio (PSNR) improvement over existing methods with even reduced complexity. The implementation of the proposed algorithm is very simple and therefore it can be readily applied to real-time video applications running on various consumer electronic products such as mobile devices. © 2006 IEEE.