Study On Image Inpainting Algorithms Based On Structure And Texture Synthesis

This thesis presents some new image restoration models based on the texturesynthesis block matching algorithm with the advantages of the structure image inpainting.Chapter1briefly introduces the research significance, the development history and thestatus of the digital image restoration.Chapter2states the background and history of texture synthesis algorithms, andintroduces some common used texture synthesis algorithms, especially Criminisi algorithm.Chapter3gives a weighted matching image restoration algorithm based on themodified priority by improving the priority setting and the matching block selection in theCriminisi texture synthesis algorithm. Firstly, the influence of the significant structure ofthe image is taken into account in the priority of the new algorithm, which makes themodified priority more accurate. Secondly, the matching block in the new method isendowed with a more reasonable weighted matching factor. Finally, the image is restoredby the new method based on the image segmentation according to the similar characteristicsof the image structure. The experimental results show that the algorithm introduced in thischapter not only avoids the bias diffusion may arise in the extant image restorationalgorithms, which makes the restored image more consistent to human being’s visualeffects, but also reduces greatly the time consumed.Chapter4presents an image restoration algorithm by integrating structure and texturefeatures with reconstruction of significant structure of images. Firstly, the algorithm splitsthe fine damaged structures and large damaged areas in an image using morphologicaloperators. Secondly, the rapid structural restoration algorithm is used for image inpainting.Thirdly, significant structures in the image are reconstructed by interpolation. Finally, thelarge damaged areas are restored by the weighted matching image restoration algorithmbased on the modified priority. For the damaged image with significant structure andtexture, the experimental results show the better restoration result and less time consumingof the proposed algorithm than that of the traditional algorithms.Chapter5does some exploration in self-acting reconstruction of the significant edgefeatures in an image, solves initially the problem of the typical VS-off marks in an image.In addition, we improve the image restoration method by using of the idea of scalingauxiliary of an image. The experiment results demonstrate the valid of the above methods.Chapter6gives a conclusion of the thesis and presents some thinking and expectationabout the image restoration.