Video super resolution enhancement based on two-stage 3D convolution
Xu Wen, Zhaohui Meng · 2021
In recent years, computer vision has penetrated into all aspects of social production and life, especially the results achieved in image and video super-resolution processing are even more amazing. The speed and accuracy of image processing based on deep learning have reached a very good effect. Although video super-resolution is composed of multiple frames of pictures. But the difference between them is that the video has the characteristics of spatial and temporal continuity. The image processing method Processing video will cut the continuity between video frames. Because medical imaging [10], satellite imaging [11], surveillance [12] and user visual experience put forward higher requirements for video resolution. This article proposes a new solution for the video from low-quality original frames to generate high-quality super-resolution generated frames. The network structure increases the attention of neighboring HR frames, extracts basic network features, and optimizes the loss function.