Real-time compression artifact reduction via robust nonlinear filtering

Low complexity postprocessing algorithms to reduce compression artifacts by using a robust nonlinear filtering approach and a table lookup method are investigated in this research. We first formulate the compression artifact reduction problem as a robust estimation problem. Under this framework, an enhanced image can be obtained by minimizing a cost function that accounts for the image smoothness as well as the image fidelity constraints. However, unlike traditional methods that adopt a gradient descent method to search for the optimal solution, we determine the approximate solution via the evaluation of a set of nonlinear cost functions. This nonlinear filtering process is performed to reduce the computational complexity of the postprocessing operation so that it can be implemented in real time. In the case of video postprocessing, a table lookup method is adopted to further reduce the complexity. The proposed approach is generic and flexible. It can be applied to different compression schemes with minor finetuning. We have tested the developed algorithm on several compressed video or images which are obtained by JPEG 2000 VM, and H.263+. It has been demonstrated that the proposed method can reduce compression artifacts efficiently with a low computational complexity.