Research On Digital Image Restoration Algorithm Based On Inpainting

The digital image inpainting technology takes a major branch in digital image processing, and also a reseach hot spot of computer graphics and computer vision. It is widely used in such domains as restoration of digital image; restoration lost information during wrieless transmission; restoration and stunts production of film; image reducion and enlargment; removing unnecessary goals, such as wrods or objects; information transmission and image compression; processing with mosaic and so on.At present, digital image repair technology research mainly on two aspects:based on partial differential model of digital image restoration techniques and texture synthesis based on the digital image restoration techniques. This paper introduces the technology of the digital image restoration of the present research situation and research significance, compared the application of a wide range of digital image restoration algorithm the advantages and disadvantages of the scope of application, and on top of this made improvement:1、Proposed an improved algorithm based on PDEThe digital image inpainting used variation PDE model construct a diffusion equation for each pixel in the missing region that relates the color expected at the pixel and in its neighbor, solving these equations, the colors surrounding the region are diffused into the region, As a general image restoration, in essence, it is belong to image interpolation.This ariticle analysis the image inpainting alogrithm based on TV model for the small damaged image repair problem, the weights of all pixels are1in the iterative calculation of TV inpainting, information of the areas which are not damaged can not spread as soon as possible in iterative calculation and it affect the speed of inpainting. We endow the pixels of different regions with different weights in order to solve that information diffusion is not fast enough, so that the information diffusion is more accurate.2、Proposed an improved algorithm based on sampleCriminisi alogrithm is the classical alogrithm in the texture synthesis inpainting technology based on sample, it is fill in the missing region by minimizing the overall discontinuity within the region for a given set of patches. Based on Criminisi alogrithm, this paper has made sevral improvements to its existed insuffciencies:(1)This paper moddify the priority function by redefining the data term to improve some deficiency in calculation the priority of patchs in Criminisi alogrithm. We take into account not only the stucture information of image, but also considerd image texture reature information, at the same time also won’t appear in the confidence value drops rapidly to zero as the filling process proceeds, priority function has a high priority of accuracy.(2)We determined the matching region acoording to the structure and texture around the patch to be inpainted, with part of the search instead of global search, it not only can save running time, but also avoid miss match, and increase the correlation between the optimal match and the surrounding neighborhood, the repair resulting more accord with the requirement of the hunman eye.(3)Make the updated confidence inverse to accumulative error when updating the confidence term, in order to avoid continuous error and lead to the phenomenon " junk piece of". The experimental results show that the improved algorithm can better applied to the image edge and complex texture repaie. This method matching result more accurate, effectively reduce the production of harmonious piece.