Cartoon And Texture Decomposition Based Biggish Region Image Inpainting Algorithm

Image inpainting refers to the process of restoring missing or damaged regions in an image. Image inpainting can be mainly classified into diffusion based image inpainting algorithms and exemplar based inpainting algorithms. Each kind of algorithm has its own applicability. Diffusion based image inpainting algorithms achieve better results on small region damaged images or cartoon images, exemplar based inpainting algorithms achieve better result on texture images. Natural images usually contain complex cartoons and textures at the same time, so a hybrid method is presented to inpaint images. Firstly, a damaged image is decomposed into the cartoon layer and the texture layer, then they are inpainted separately, and the two inpainted images are added together as the final inpainting result. The main works of this thesis are as follows:(1)An automatic estimation algorithm of shear operator’s shear angle and shear direction in the cartoon and texture decomposition algorithm is presented.The angle and direction parameters of the shear operator in the existing algorithm are fixed.The cartoon and texture are separated incompletely. An automatic estimation algorithm which is based on histogram of gradient of those parameters is presented. The experiment results show that the improved algorithm can not only improve the effect of cartoon and texture decomposition but also the value of PSNR.(2) A cartoon layer inpainting algorithm combined with edge construction and super resolution is presented.Blurring effect is introduced while applying cartoon layer inpainting algorithm to the biggish region damaged image. For shrinking the area of the damaged region, damaged images are separated into two categories according to if there is an edge in the image and then are dealt with separately. When the edge exists in the damaged region, an edge reconstruction method is used to shrink the damaged region, and fast marching method is used to inpaint the shrinked damaged region. When there are no edges in all the connected domains of damaged regions, down sampling method is firstly used to shrink the damaged region, fast marching method is used to inpaint the shrinked damaged region, and then a super-resolution method is used to reconstruct the image. The experimental result shows that the improved algorithm can not only weaken the blurring effect but also improve the value of PSNR.(3) A geometry distance based texture layer inpainting algorithm is presented.For the inaccuracy of the priority and the deficiency of the matching strategy which cause the undesirability of the inpainting results, a improved algorithm is presented. In the repairing order aspect, the priority model is modified, in the matching strategy aspect; Euclidean distance is combined with geometry distance. Multi-block repairing strategy is adopted to confirm the best matching blocks of every block.The experiment results show that the proposed method overcomes the deficiency of the constraint of the matching strategy and the drawback of the incorrect texture propagation. It has better visual effects compared with the state of the art algorithms.The experiment results show that the proposed method can preserve the structure and keep the propagation of the texture at the same time, and it also achieves better performance on visual quality.