Low Bit-rate Image Compression based on Post-processing with Grouped Residual Dense Network

Seunghyun Cho; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

In this paper, an image compression method implemented for CVPR 2019 Challenge on Learned Image Compression (CLIC) is introduced. It is designed to satisfy both requirements of image compression, "higher compression ratio" and "better quality", at the same time. To this end, a neural network based image quality enhancement is incorporated into the most recent traditional image/video coding technique. The decoders, ETRIDGU, ETRIDGUlite, and ETRIDGUfast, which implement the proposed image compression method are designed to have different degrees of complexity and compression efficiency. ETRIDGU, which provides the highest compression efficiency, is reported to achieve the 2nd highest PSNR in the lowrate track of CLIC. ETRIDGUlite, which compromises between the compression efficiency and the complexity, is reported to be the fastest one among the decoders with high mean opinion score (MOS) in the same track.