Research On The Algorithms Of Digital Image Inpainting

Digital image inpainting is an important research direction in the field of digital image processing and the technology has a high application value, which can be used to protect the cultural relics, repair the damaged photos, hide away the video error, ma ke the film and television play’s special-effects and remove unnecessary objects in the image, etc.. At present, the digital image inpainting algorithm are divided into two kinds:the digital image restoration algorithm based on the texture and the digital image restoration algorithm based on texture synthesis. This paper analyzes the advantages and disadvantages of these two kinds of algorithms, and carried out a thorough research from the following three aspects.Firstly, Small area damaged image inpainting problems. The origina l total variation(TV) model using the anisotropic diffus ion only and it is sensitive to the selection of parameters, etc. To overcome these shortcoming, an improved TV model of image restoration algorithm which based on adaptive iterative equation and iterative rounds of diminishing was proposed. The proposed algorithm using the difference of frequency of the regions that need repaired to adaptively select iterative equation. In the early stage of iterations, in order to speed up the iteration, the parameters and the iterative stopping threshold are set to greater value; in the late iterations, the parameters and the iterative stopping threshold value are adaptively determined to keep sharp edge information. The results of experiments showed that the improved algorithm can improve the quality of the repair by adaptively achieving the combination of isotropic and anisotropic diffusion in the image restoration, and effectively avoid ing the step effect.Secondly, large area damaged image inpainting problems. In the Criminis i algorithm, the boundary priority calculation formula was easily influenced by confidence item value reduced to zero, and the template window size is fixed in the process of search. To solve the problem, an adaptive Criminis i algorithm based on gradient information was proposed. In order to optimize calculation formula of the priorities and template size, the proposed algorithm using the gradient information of the area to be repaired and Introduce Parameter to represent the weight of the confidence item and the data item, respectively. The results showed that the improved algorithm can keep the continuity of the structure and texture information of the image, and effectively improve the quality and efficiency of the repair.Finally, the image decomposition technology are used to extract the structure information and texture information from the image. The improved TV model are applied to inpainting s tructure part of image, and the improved Criminisi algorithm is applied to restore texture part of image. Then the result image was formed by the fusion of the structure image and the texture image. The simulation results showed that, compared with using TV model algorithm, Criminis i algorithm and the original algorithm separately, the proposed algorithm possess better repair effect.