Enhancement Of Screen Content Video Based On Deep Learning Scheme

In recent years,with the rise of various e-learning websites and e-sports live broadcasting industry,the special type of screen content video is gradually inseparable from people's life.Therefore,how to effectively improve the quality of the screen content video has become an urgent problem to be solved.The screen content video mainly includes various combinations of text,patterns and part of the traditional video images generated by electronic equipment.Video games,animations,and slideshows and documents as we know all belong to the category of screen content video.The video captured by the camera equipment will have very obvious noise caused by the sensor.The continuity between the video frames is great,and the color space ranges widely.The screen content video is basically all generated by electronic equipment,so there is no sensor noise.In addition,the color space of the screen content video is very limited,in company with a large number of similar color areas.At the same time,the saturation of the screen content video is very high,and the color blocks in the video are mostly uniform and flat.After compression and encoding,the quality of the screen content video will drop sharply,and the video will produce many compression artifacts in the visual effect.As a result,this thesis mostly focuses on how to improve the quality of the compressed screen content video.The research content of this thesis can be viewed below:1.This thesis proposes a cross-fusion module to remove video artifacts as much as possible to improve video viewing quality.At present,there are few researches on the quality enhancement of screen content video.The mainstream compressed-video-qualityenhancement-schemes do not come up with a targeted manner towards the screen content video.Especially,a great deal of frame alignment methods cannot be well adapted to the screen content video.Therefore,the cross-fusion module we proposed does not use an alignment method,but it enhances the quality of compressed screen content video through the feature fusion scheme.2.This thesis proposes an enhanced framework for efficient adaptation of screen content.Aiming at the difference between the screen content video and the traditional video,a block classification scheme is used to enhance the screen content video.Firstly,we divide the input video frame,and then the image blocks are divided into two categories according to the ease of restoration and reconstruction.Finally,different kinds of image blocks are enhanced by different networks.In this way,we can drastically speed up the model's processing of video with limited quality loss.