Image Inpainting Based On Adaptive Anisotrope Diffusion

Image inpainting is becoming a hot topic in computer vision research. This paper mainly studied image inpainting algorithms based on structure, which used to repair smaller damaged region of the image. Image inpainting algorithms based on structure mainly use the local image structure feature, by establishing some rules, the image information can diffusion along the direction of isophotes. The processes of information diffusion can be reflected through diffusion equations. Therefore, anisotropic diffusion is the basis of image inpainting based on the structure. However, the anisotropic diffusion equation has some shortcomings such as sensitivity to noise, slower diffusion rate, missed corner information and so on. In order to overcome above problems, some works has been done in this paper:Firstly, this paper improved TV models. The improved model called double crossed TV changed the TV model from 4-neighborhood to 8-neighborhood, and this idea can be found in later works. This approach can effectively improve the accuracy of digital image inpainting.Secondly, this paper designed an adaptive diffusion model based on damaged region classification. This model can distinguish less obvious structure region and the obvious structure region. And in different regions, the model can diffuse in the corresponding ways. The model solves the corner information in degenerate diffusion. The improved model solved the problem of missed corner information in the degenerated diffusion model and because of the new diffusion rate function, diffusion rate is increased.Then, this paper designed an adaptive diffusion model based on local feature classification. Average gray matrix model was introduced in this model to measure the local features of the image. Besides, this paper established a diffusion model related to the average gray matrix, so that diffusion direction and diffusion rate are properly controlled.