Some Researches On Image Disocclusion Based On Level Sets

The researches on image processing based on Partial Differential Equation (P.D.E.) began in the 1960s' and 1970s', P.D.E. was introduced for the perspective of noise removal and image restoration. However, it was not until 1990s' that P.D.E. was systematicly introduced to the field of image processing, and formed a relatively integrated theory system combined with other mathematic tools such as mathematic morphology and affine geometry. The level set methods proposed by S. Osher have a good property of geometry interpolation, and gray interpolation can be changed into curve interpolation. Actually, it has great influence on image processing and have achieved satisfactory results in image restoration, noise-removing, edge detection, image matching, and image recognition.This thesis focuses on the applications of level set methods in image restoration. In the light of the image damage caused by occlusion or part stain, first we extract the level sets of the image and then through the analysis of these level sets to fill up the occlusion with level set, thus realize the restoration of image. In this paper, the occlusion is selected artificially and meet a certain geometry conditions, this is because in the practical application the occlusion is distinctive and can be determined by a few physical and chemical methods of file protection. After the selection of the occlusion, we go on to determine its frontier, the curve of which has a certain geometry characters. What we pay much attention is the so-called T junction. After finding the T junctions on frontier, we try to match these T junctions with each other. We can connect the T junctions with level lines after the matching up of T junctions. The connection of all level lines means that the broken level lines in the occlusion has been filled up, that means the image has been restored. When connecting level lines, we use the minimum energy as the optimization standard. On the basis of it, we use geodesic curve as level lines to connect the matched-up T junctions. Meanwhile we also use meaningful beeline detecting methods to deal with unexpected condition in processing the frontier of occlusion.Finally, through experiments of figure image and scenery image we proved the method mentioned in the paper has a satisfactory effect on image restoration and can be well applied in file management and old photo preservation.