Research On Digital Image Inpainting

Some information of images will be lost because of saving for a long time or other reasons. Digital image inpainting is to fill the lost information, and make sure that the image will not be easily detected of having lost after inpainting. Image inpainting technology has a high application value in art restoration,object removal,lossy image compression and error concealment in video communication. In this paper, inpainting algorithm was researched and the existing methods were improved so that the restoring effect becomes more perfect and they can be more suitable for practical applications.The three key points of exemplar-based inpainting methods——calculation of priority, size of window and ways of searching the best match block were researched and improved. Most algorithm don not take the continuity of the structure into account, so an exemplar-based inpainting algorithm constrained by continuity is proposed in this paper, it can overcome the disconnecting phenomenon when filling the strong structure image. According to the influence of window size had on texture restoration,a new adaptive window size inpainting algorithm was presented to avoid the distortion caused by improper window size. Similar textures are also centrality,so local searching method was presented to avoid introducing irrelevant information when using global searching.Fast image inpainting algorithm was researched. Telea's algorithm can not well preserve the image boundary and it can also cause error accumulation, so a fast and efficient algorithm was proposed by introducing a reasonable assessment method of pixel position and confidence factor, combined with Bornemann's algorithm. The results show that this algorithm effectively improved the inpainting quality, within the acceptable range of time.Finally,image inpainting was applied to error concealment and image enlarge in this paper, to explore its application in other fields. For error concealment, inpainting methods not only had better results, but also can be used in any irregular lost information. For image enlarge, inpainting model can avoid the sawtooth effect which traditional methods have because it is PDE methods and it reconstructs the image information smoothly.