Fast Best Neighborhood Matching algorithm for intra block error concealment in H.264/AVC

In this paper, the Best Neighborhood Matching (BNM) error concealment algorithm for images is extended and optimized to intra block in H.264/AVC video decoder. Based on the block-wise similarity within the block-coded images, BNM method utilizes the information of not only neighboring pixels nearby the range block, but also remote regions in the picture so that this method can get the better reconstructed image. However, the whole searching process for the best domain block has to cost too much time to implementation in the real-time video decoder. Hence, two optimal strategies are provided here to reduce the computation complexity in the searching process. Successive elimination principle is applied by comparing the Mean Absolute Difference (MAD) between the range block and the domain blocks by analyzing the inequality relation. The basic idea is that obtaining a criterion as the necessary condition of the best matching domain block, by which some impossible domain blocks can be eliminated directly without searching process. The other strategy is that the radius of search window can be adjusted adaptively by analyzing the characteristic of the block in motion and texture information. Simulation results in the H.264/AVC video decoder demonstrate that the performance of the proposed algorithm decreases intensive computations effectively compared with the original BNM method, and better recovery results are gotten than the conventional interpolation approaches.