SR-LBSCC: Super resolution based screen content image compression at low bitrate

Abstract

The massive amount of screen content data puts enormous pressure on storage and bandwidth, and traditional screen content encoding methods (e.g. HEVC-SCC) cannot meet the requirements of high efficient compression. The main challenge is how to retain the sharp edges when screen content images are encoded with low bitrates. In this paper, we propose a super resolution (SR) based screen content image compression method for low bitrate encoding (SR-LBSCC). Concretely, first, the original image is down-sampled and compressed with the traditional screen content coding method, which significantly reduces the encoded bitrate. Then, a novel SR neural network is designed to enhance quality of reconstructed image, which could recover details and remove compression artifacts. Finally, experimental results show that compared with the standard HEVC-SCC, the proposed method could averagely reduce bitrate cost by 17.02% and 24.97% respectively on the SCID and SIQAD datasets at similar subjective quality.