Deep Learning Based Video Super-Resolution and its Application in Video Conferences

As a remote communication method, remote video scene is widely used in some occasions such as telecommuting, telelearning, and teleconsultation. However, the remote video scene requires a large network bandwidth for image information transmission, resulting in the insufficient real-time performance. In this paper, applying the super-resolution image restoration method to a remote video scene only requires 1/16 of the original network bandwidth, but a good visual effect of image reconstruction can be obtained. The Enhanced Information Multi-Distillation Block (EIMDB) and the Pixel-Level Information Distillation Block (PIDB) are proposed, which can improve the super-resolution effect of the image with a small amount of calculation. Finally, a novel real-time remote video communication super-resolution network (RVCSRN) is proposed which achieves a good balance between speed and restoration effect, and can effectively improve the visual effect of the remote video scene. In addition, since the pictures processed in the remote video scene are different from those processed by the general super-resolution method, these pictures have a compression loss, so a remote video scene dataset (RVSet) is created to obtain a better super-resolution effect.