Research And Implementation Of Video Super-resolution System Based On Neural Network

In recent years,with the improvement of hardware equipment and computing power,neural networks and deep learning technologies have developed rapidly.Meanwhile,traditional image and video processing techniques have also been greatly developed through neural networks.Among them,image and video superresolution technology has always been a hot research topic in the field of computer vision.At present,most online video websites will provide a category,which provides a function of switching multiple video resolutions for different users and different network conditions.The principle is to push video files of different resolutions according to the user’s network conditions.However,in real life,limited by the shooting equipment,shooting conditions and transmission conditions of the images,users often cannot obtain images and videos that meet the expected resolution.At this time,the resolution of image resources can be increased at the software level by using the super-resolution technology of video.Based on the above requirements,this paper proposes a video super-resolution method based on adversarial neural network,and based on the analysis of system requirements,realizes the corresponding video super-resolution system.The main research and implementation contents are as follows:(1)An improved GAN-based video super-resolution algorithm is proposed.The temporal information in the video is introduced through a recurrent backprojection network to build a generator;and a discriminator is built through improved super-resolution.Two networks construct an adversarial generative network.After training,save the model file and parameters to provide superresolution functions for the system.(2)Realize the video super-resolution system,provide super-resolution services for videos,and implement the video super-resolution algorithm.Users can upload videos through the system,use the super-resolution model to obtain high-definition videos,and download videos.This paper designs a GAN-based video super-resolution algorithm and conducts experiments on public datasets.The results show that the algorithm achieves better results on videos than previous ones.At the same time,this paper builds a video super-resolution system and implements five types of models including the algorithm in this paper.A test plan for corresponding functions is proposed,and the system and model are quantitatively evaluated in terms of performance and video quality.The results show that they are in line with expectations.