Coding Artifact Reduction Based on Local Entropy Analysis

Coding artifacts are very annoying in highly compressed images and video sequences. Most artifact reduction techniques blur the details of the images while removing various coding artifacts. In this paper, we propose a novel and explicit approach for classifying blocks into detailed regions, intermediate regions and smooth regions. The classification is based on the information content of the underlying region. The information content of a region is quantized by local entropy, which is calculated on the PDF of the pixel intensity distribution. Local entropy is used as an indicator of how much smoothing is needed for a certain region. It is well known that blocking artifacts are more visible in flat regions than in detailed regions. We apply mild low-pass filters on detailed regions to preserve the sharpness, and strong low-pass filters on flat regions to remove the severe blocking artifacts. Experimental results show that our proposed algorithm can preserve the details and reduce coding artifacts better than more expensive state of the art techniques.