Shifted Window Based Filtering for Alleviating Blocking Artifacts

We propose a computationally simple yet effective image/video deblocking algorithm based on block level postfiltering. Each image block is shifted in the range of half of the block dimension to find the candidate blocks for smoothing. The Mean Absolute Error (MAE) between the central image block and the block indicated by a shifted window is used as a criterion to determine if the corresponding shifted block is used in smoothing. The threshold of the MAE is defined with analyzing the quantization table used in the codec. The proposed algorithm successfully removes the blockiness while preserving the edge features. Another merit of the proposed algorithm is that it is applicable to almost all bit rates without the requirement of detection on blockiness position or estimation on blockiness severity, thus does not impose computational overhead. The experimental results demonstrate that in a wide range of bit rates, the processed image retains higher PSNR and better visual quality.