A new post-filtering technique for block-based transform coded images
Shan Suthaharan, Hong Ren Wu, Michael Yuen · 2002
The two-dimensional discrete cosine transform (DCT) is probably the most popular transform in image and video coding systems as it achieves very high compression ratio for fast implementation. Various still digital image and video compression algorithms using this block-based transform coding method cause the artifacts, called the blocking, in the reconstructed images of a video sequence. This artifact is more prominent at low-bit rates. A new eight-neighbour symmetric post-filtering technique is introduced in this paper to reduce the widely known blocking effect which is commonly found in the reconstructed images of block-based transform coding schemes. In this technique, the weights of the eight neighbouring pixels of the pixel of interest are determined by the modified Lagrange's interpolation algorithm. The proposed technique reduces the blocking effect while maintaining the sharpness of the reconstructed image. The results obtained by the proposed technique are compared in this paper with the results obtained by several well known eight neighbour low-pass filters such as Reeve and Lim's method, the Laplacian eight neighbour accounting filter (LENAF), the Laplacian four neighbour filter with + form, the Laplacian four neighbour filter with x form and a two neighbour filtering method. These comparisons will be conducted with the following measures: peak signal-to-noise ratio (PSNR), block-edge impairment metric and subjective quality evaluation.