A video compression artifact reduction approach combined with quantization parameters estimation

High Efficiency Video Coding is one of the most widely used Video Coding standards. It could encode videos to bitstream with a high compression rate for transporting, and videos would be reconstructed by decoding bitstream. However, decoded videos with compression artifacts would degrade the viewing experience of users. At present, many methods based on deep learning have been proposed to reduce compression artifacts for quality enhancement. Most methods train their models assuming that quantization parameters (QP) are known. But the QPs of videos are obtained from bitstream, which means QPs are probably unknown if there are only decoded videos. Without QP, training a network for compression artifacts reduction would become difficult. To this end, we propose a video compression artifacts reduction approach combined with quantization parameters estimation (VRQPE). In light of the fact that decoded videos with different QPs have different artifacts, we extract representative sample blocks in video frames to estimate the QPs of these videos. Then, we propose a majority voting to achieve better estimation results. Afterward, we proposed a quality enhancement network to reduce compressed artifacts. Extensive experiments demonstrate that our proposed VRQPE achieves superior performance.