In this work we propose a novel postprocessing technique forcompression-artifact reduction. Our approach is based on posing this task as aninverse problem, with a regularization that leverages on existingstate-of-the-art image denoising algorithms. We rely on the recently proposedPlug-and-Play Prior framework, suggesting the solution of general inverseproblems via Alternating Direction Method of Multipliers (ADMM), leading to asequence of Gaussian denoising steps. A key feature in our scheme is alinearization of the compression-decompression process, so as to get aformulation that can be optimized. In addition, we supply a thorough analysisof this linear approximation for several basic compression procedures. Theproposed method is suitable for diverse compression techniques that rely ontransform coding. Specifically, we demonstrate impressive gains in imagequality for several leading compression methods - JPEG, JPEG2000, and HEVC.
展开▼
