An Image Inpainting Algorithm Combined With Isophotes And Exemplar

In the existing exemplar-based image inpainting algorithms, there are obvious repair traces and false match when the large area is repaired. To overcome this problem, this thesis focuses on the research and improvement of exemplar-based algorithms to improve the repaired result.In this thesis, the research background,significance and the development history of digital image inpainting algorithms are first introduced. Then, some kinds of schemes, such as several classical modal based on PDE algorithm,the image decomposition inpainting algorithm and the Criminisi algorithm based on exemplar are described.In the Criminisi algorithm, the inpainting result can be improved because of the fixed patch size. As a result, this thesis introduces an adaptive size of patch algorithm for inpainting. The PDE algorithm is suitable for repairing the quite small damaged area and the exemplar-based algorithm is suitable for the quite big damaged area. Based on this character, the Isophote-Criminisi algorithm is put forward. Substantially, the algorithm brings in a filter operator to divide the damaged region into two parts and different parts are repaired with different algorithms. The experimental results show the feasibility and effectiveness of the divided region algorithm for inpainting.The method of priority calculation in Criminisi algorithm may cause the inpainting order along a line of the structure. In order to solve this problem, the similarity of inpainting patch and neighborhood field is used to determine the priority. The experiments proof effectiveness of the improved algorithm for structure image. On the other hand, the mtheod of choosing the best matching patch may cause the error match phenomenon. Then, the histogram factor is brought in to choose the best matching patch. In order to reduce the patch effect, this thesis uses three patches to determine the inpainting patch. And then experiment results proof effectiveness of the improved matching patch algorithm. At last through masses of experiments show that the improved priority and match patch algorithm for inpainting is effective.