Study On Image Inpainting Alogrithms

Inpainting can originally be looked backward into Renaissance when it was a method for artists in museums to bring the medieval pictures"up to date"in order to make them more legible and to restore them unity. With the fast development of the computer , the methods and technologies of inpainting have been improved quickly. After a long period of observation in the museum, Bertalmio etc. first proposed the term"digital inpainting"in the July of 2000 and set up partial differential equation to solve the problem. It was a breakthrough which replaced the manual work of artists. This technology saved much time and at the same time made the inpainting result modified. In the recent years, due to deeper research on digital inpainting many new methods and technologies have appeared which have been applied in more and more fields.In this thesis, the research process of image inpainting is devided into two steps: selecting the inpainting domain and designing the methods to inpaint, for which new techniques are approached. The major contributions of the thesis are presented as follows:1 The semiautomatic method is chosen to select the inpainting domain and is presented by the design flow chart. First, seed pixels are selected for region growing. Second, region merging goes on which is completed by computer. If the result of inpainting domain is not desirable, manual method should be selected. In the semiautomatic method, region growing, region merging and human-computer alternation are combined to deal with relatively more complex images.2 The inpainting method based on skeletonizing is proposed. At the same time of obtaining the skeleton of the region, interpolation should be completed by the usage of the surround information. At last only the skeleton is left and it is interpolated by information of the finished part.3 Reverse distance weighted method is applied into image inpainting. It is a geometrical method originally used in image interpolation. The designing idea is based on the principle"The shorter the distance is, the more the contribution for inpainting is ."4 The applications of nonlinear diffusion equation are extended. They are not only involved in image enhancement, image segmentation and image texture analysis, but also in image inpainting. Diffusion is a problem of thermodynamics, but it has more applications in image processing. Via the improvement of anisotropic diffusion equation, the method is fit for image inpainting.5 Fast digital image inpainting is improved. Via selecting the non-symmetric convolving kernel, the inpainting result can be better.