Postprocessing Algorithm in DWT-Based Image Compression by Using Multiscale Adaptive 2D Filtering

In DWT-based image compression, as the coding rate becomes low, the quality of the coded image degrades severely due to the quantization of wavelet coefficients and the removal of high-frequency wavelet coefficients. Especially, in the vicinity of edges, the ringing effect inevitably comes up. In this paper, a multiscale 2D filtering method based on local primary direction is investigated which may significantly improve the quality of the image compressed. By the method, local primary direction is detected first, according to which the image was filtered adaptively. To ensure the robustness of the detection of the local primary direction, a circular shift method is adopted. To remove the high-frequency noises in all the bands, the decoded image is enhanced from coarse to fine scales. The simulation results illustrate that the method proposed may improve the quality and the visual effect of the compressed images.