Adaptive MLP post-processing for block-based coded images

Block-based image coding techniques are widely used for encoding images and videos. However, many annoying artefacts appear when an image is encoded at low bit rates. In these artefacts, the blocking effects are very obvious to human vision. Thus, an efficient blocking effect reduction scheme is essential for preserving the visual quality of decompressed images. A new adaptive post-processing algorithm is proposed to reduce the blocking artefacts of block-based coded images by using neural network techniques in the spatial domain. The algorithm combines a variance-based classifier and multilayer perceptrons to improve the performance of post-processing. In the proposed algorithm, the blocking and ringing effects in a reconstructed image are diminished without blurring of the edges, and the detailed region in the image is also enhanced. Comparison results between the proposed algorithm and other algorithms are made with several Joint Photographic Experts Group and vector quantisation decompressed images. In the simulations, the results of reconstructed images with improvements in both visual quality and PSNR are shown. It is found that the proposed algorithm is an effective post-processing algorithm for block-based image coding at low bit rates