CESSING OIF T FILTERING AN

Stephen Cheong Yu Chan, Tian-Tsong Ng · 1997

Transform coding often leads to artifacts called blocking when the image or videci is compressed to low bit-rate. In this paper, a new post-processing scheme using selective low-pass filtering and fuzzybased enhancement is proposed. The merit of the scheme is that it can remove most of the blocking artifacts found in image or video coding while preserving most of the fine details. Simulation results demonstrated that the proposed algorithm can significantly improve the visual quality of the decoded image. I. 1NTRODUCTIC)N Transform coding using the Discrete Cosine Transform (DCT) is the most popular approach for image compression. However, it often Yeads to artifacts called blocking when the image or video is compressed to low bit-rate. This artifact can be reduced by proper selecting the quantization step sizes in the coder andor by post-processing the decoded image. Blocking artifact originates from the block: based nature of transform coding. Since the image is divided into square blocks and each block is ccided and quantized individually, there will be noticeable discontinuity between adjacent blocks when the picture is encoded at low bit-rate. Since blocking is a high frequency artifacts, low-pass filtering can be used to reduce its effect. However, simple low-pass filtering usually leads to over-smoothing of the high frequency components found in edges and textures. Previous research on post-processing using spacevariant filters [l] was done by smoothing of the edges and by estimating the edge irlfonnation in the compressed image data [2]. This is a rather difficult in low bit-rate coding since the actual edge information will be severely distorted. Other approach based on iterative method has also been proposed. In [3], the image is reconstructed using the method of projections onto convex sets (POCS) [3] where a number of constrains on the coded images such quantization stepsize are used together with low-pass filtering to remove the blocking artifacts.

Read the paper · More papers on PaperTik