Content-adaptive Resolution E nhancement of Compressed Video with Encoder-Generated Side Information

In this paper we propose a technique to dynamically generate optimized filters for the resolution enhancement of compressed video. The filters are derived during the encoding and sent to the decoder for post-processing. By making use of the scene-level and pixel-level classification, we adapt the filters to both the global and local content. Our experiments show 1-2 dB gain in PSNR as compared to the state-of-the-art techniques with ignorable overhead. Furthermore, it also improves the subjective quality with sharper images and suppressed artifacts.