Research On Bayesian Matting Algorithm And Application In Video Image Matting

The matting is a technique that uses an effective algorithm to completely separate the foreground objects of interest in the image from the background area.The matting is the basic content of digital image processing and computer vision,and has a wide range of applications in photo editing,video production,virtual reality,television media,advertising and other fields.Commonly used matting algorithms mainly include Knockout diagram method,Poisson diagram method,Bayesian diagram method,and Hillman diagram method.The basic implementation steps are as follows: manually select a foreground object in an image with a closed outline.The interior of the contour line is the foreground area.The exterior of the contour line is used as the background area.The area between the outline line and the edge of the foreground object serves as the middle area.The foreground area,background area,and middle area form the initial three-dimension map;The point in the middle region belongs to the foreground or the background.After that,the foreground object is completely separated from the entire image to achieve the purpose of front and background separation.Through the image synthesis technology,the obtained foreground image is fused realistically and seamlessly.In another image,a specific effect is obtained.This paper focuses on the Bayesian graphing algorithm and its application in video mapping.Firstly,the Bayesian graph algorithm principle is analyzed.Second,for the problem that the three-part graph needs to be manually selected,the construction method of the three-dimension map is studied.The region growing method is used to extract the contour of the foreground object region,and then Bayesian algorithm is used to do the map calculation to improve the accuracy of the three-dimension map.Then,the Bayesian map algorithm is applied to the video map.The three-dimensional map needs to be calculated for each frame of the video.For the problem of low efficiency of graphs,the three-dimension map of key frames is constructed first,then the bipartite optical flow interpolation method is used to automatically generate the three-dimension maps of the middle frames,and then the three-dimensional map sequence of the entire video frame is obtained,thereby realizing the Bayesian video.Finally,the effectiveness of the proposed algorithm is verified by experiments.