Research And Implementation Of Video Alignment And PatchMatch Algorithm In Video Inpainting

Video repair(video inpainting)refers to the process of replacing or reconstructing video information containing lost or damaged parts.by using appropriate repair methods and using known video frame information to repair missing areas,the repaired video has this pleasant viewing effect visually.Visual information is the main source of people's access to information,visual information includes picture information,text information and video information.Because of the convenience and appreciation of video information,it has become the mainstream way of information carrier.Therefore,the problem of symbiotic video repair is also important.In all areas of life,video repair plays a pivotal role,such as repairing damaged old movies,removing obstacles or people in video,changing video stories,etc.And in the video repair process,one of the most important work is the alignment operation and image block matching(PatchMatch)process.alignment accuracy is the cornerstone of the subsequent repair process,and whether to match to the correct image block directly determines the final repair effect.It's well known that when the camera motion is rotated around the fixed point or the scene is flat or the motion range is small,the camera motion can be described by global Homography.however,when the camera motion does not meet these limitations,using the global Homography approximation camera motion brings large errors,which directly affects the subsequent video repair effect.For image block matching algorithms,such as Brate force algorithms,they are usually enumerated and exhausted.The algorithm is simple but time-consuming,while the target graph to be repaired can only be used in motion consistency-based methods searching in a single image;in addition,the general optimization-based methods only accept better solutions than the current value in the search and iteration process,and easily fall into the local minimum point.In view of the above problems in the process of video repair,this paper mainly does the following three parts of work:(1)A dynamic video repair method based on local model alignment is proposed.for a large range of camera motion,this paper uses local Homography to align video frames,where the alignment accuracy of local model is closely related to the number and distribution of feature points.so this paper combines a corresponding point matching method based on motion consistency in local model estimation.this algorithm greatly increases the number of feature points matching and thus improves the alignment accuracy of the algorithm.The experimental results show that the proposed local model-based alignment method improves the video alignment accuracy,thus effectively improving the video repair performance,expanding the scope of application of traditional video repair methods.(2)A PatchMatch algorithm based on improved bat(Bat Algorithm,BA)is proposed.Inspired by the idea of image repair,this algorithm uses the heuristic idea to find the most similar image blocks in the current video frame or image library to fill the area to be repaired.The algorithm constructs chaotic functions and constructs a new positioning system in the traditional BA algorithm,which is extended to multiple reference maps in the search stage.Experiments show that the repair effect of this algorithm is higher than the image evaluation index of classical PatchMatch algorithm,and the visual effect is better.(3)A dynamic video repair system has been developed.The specific functions of the system include video preprocessing module,video frame alignment module,clipping module,repair module,integrated video module,changing video rate module and so on.Experiments show that the system has high reliability and strong practicability.