Research On Digital Video Restoration Algorithm Based On Structure And Texture Information

In recent years,the research on video inpainting in the field of image processing is constantly in-depth development and application.Compared with single static image inpainting,video contains more structure,texture and motion feature information.Video inpainting is the most direct and effective way to save the target information.But video inpainting calculation is more difficult and time-consuming,so it has more research value.Based on the sample block repair algorithm,this paper mainly studies the search algorithm,structure preservation,texture similarity and other related issues.The specific research contents are as follows:Firstly,the gradient information of the image is introduced into the objective function as the structure information of the video,and a method of repairing the structure constraint is proposed.The time dimension of video is used to build the temporal and spatial pyramid of video,from top to bottom,and then from outside to inside to repair the boundary pixels of damaged area;then the weighted average method is used to carry out rough reconstruction,update the weight,and then update the boundary.On the one hand to determine the structural characteristics of damaged area,on the other hand to supplement the reconstruction information;the same method is used to repair the pyramid of each layer at a time.Secondly,after the structural features are recovered,texture migration technology is used to make the damaged area image texture consistent.The stylized texture in the source code is used to synthesize the stylized image of the target and optimize the patch of the energy function;then the weighted average energy value is used to improve the iteration through the structure guidance;by voting,using the given reconstruction target stylized image,calculate the weighted average color of pixels located at the same position in adjacent patches to generate a repair image.The experimental results show that the method is effective in visual perception,PSNR and MSE indicators are better than other algorithms.Finally,for the complex video files that move both foreground and background at the same time,the fast reverse search of optical flow is used to obtain more accurate matching blocks.We use optical flow to improve the search process,and then use multi-scale aggregation to estimate the fast dense flow.Finally,we use the fast variational refinement step to further improve our solution based on the dense inverse search method.The experimental results on different data sets show that theoptical flow estimation is accurate and the image recognition effect is better than other recognition methods.This paper has 35 figures,11 tables and 85 references.