The Application And Research On Digital Image Inpainting Technology In Processing Of Damaged Pictures

The digital image inpainting technology takes a very important branch in digital image processing, it is widely used in such domains as repairing broken pictures, photes, paintings and film strips; inpainting ancient cultural relic; eliminating unnecessary goals, writings or objects; manufacturing film and television stunts; image transmission and compression; image reduction and enlargement and so on.Until now, the research and the application of the digital image inpainting technology is mainly divided into two technologies which are "the digital image inpainting based on variation PDE model" and "the digital image inpainting based on texture synthesis"."The digital image inpainting technology based on variation PDE model" uses mathematics method to establish partial differential equation and to obtain the isolux direction on the boundary of damaged image region, disseminates the effective information from the boundary to the damaged region by using proliferation principle. According to that the texture of image has the characteristic of similarity, uniformity and continuity, "The digital image inpainting technology based on texture synthesis" carries on the effective synthesis to the texture to complete the inpainting, it is mainly suitable for the image with quite rich texture and quite big damaged area. At present, this kind of technology mainly includes "the texture synthesis inpainting technology based on the image dissection" and "the texture synthesis inpainting technology based on the sample".A.Criminisi algorithm is a little classical algorithm in "The texture synthesis inpainting technology based on the sample". It is to select the picture element block with biggest inpainting priority on the boundary of damaged area, then to search in the source region outside the damaged area, seek the most similar templet plate to the current picture element block, substitute for the current picture element block, finally complete the inpainting of the damage region. Compared with other inpainting algorithms, A.Criminisi algorithm's inpainting speed is quicker, inpainting quality is better, especially inpainting on image with quite big damaged area, its merit is more prominent. Therefore, A.Criminisi algorithm is widely applied in domains such as image inpaintings, video inpaintings and so on.However, A.Criminisi algorithm also has some shortcomings such as the template size is fixedly invariable, the time spent on searching matched block is a little long, the inpainting order is easy to make mistakes, the joint trace between the block and block is quite obvious and so on. This article mainly studies A.Criminisi algorithm, and has made two improvements to its existed insufficiencies:First, whether image information is rich or not, its size of templet plate is fixed when using A.Criminisi algorithm to repair the damaged image, whose effect is not very good regarding to the processing of certain complex image. This article judges the image information to be repaired is rich or not by accounting the number of the image gradation level, according to that the image information is rich or not, completes the auto-adapted choice of the template size, if the image information is rich, then chooses the small template, if the image information is not rich, then chooses the big template.Second, when using A.Criminisi algorithm to repair the damaged image, the inpainting order is according to the formula of the inpainting priority, and the weight occupied by the structure information content D(p) does not change along with the change of inpainting process, the inpainting order determined by which is not completely reasonable, easy to make mistakes, especially for the image with quite big damaged area, whose inpainting effect is not very good. This article enhances the function of structure information content D(p) in determining the inpainting priority of the pixel which is on the boundary of damage region through introducing increasing weight operatorλ, thus obtains the better, more reasonable inpainting order, it can obtain a better inpainting effect when it repair an image with wide broken area.At the same time, this article carries on massive simulating experiments about the two proposed improvement algorithms, which confirms the feasibility and the validity of the improved methods.