Directional adaptive loop filter for video coding

In this paper, a directional adaptive loop filter is proposed for video coding. It classifies the pixels of a reconstructed frame into multiple categories based on the local directional characteristics. A Wiener filter is estimated for each category by minimizing the average mean square error between the original and reconstructed pixels in that category. The filter supports are carefully designed by exploiting the directional information in order to reduce the amount of overhead. The experimental results show promising performance in various coding configurations for both subjective and objective quality.