Improvement of Coding Efficiency Method Based on Super-Resolution by Learing Decoded Images in HEVC

Recently, high-resolution videos such as 4K and 8K have become increasingly popular, leading to a significant increase in video data size. Consequently, techniques for more efficient video compression are needed. One approach to address this issue is to downsample the video image to reduce the amount of information and then perform encoding. Upsampling is performed using super-resolution or other methods to restore the original resolution, which is expected to improve coding efficiency. However, super-resolution may not necessarily produce superior results compared to other. As conventional superresolution processing does not consider degradation caused by video encoding processing in the learning process. Therefore, we propose a method of learning super-resolution that considers the degradation caused by both downsampling and video encoding processing. As a result, we can perform super-resolution processing suitable for video encoding and improve encoding efficiency by up to 7%.