The Study On Image Inpainting Algorithms Based On Multi-Scale Geometric Analysis

Image inpainting is a research focus of computer graphics and computer vision. It has been widely applied in many areas, such as cultural relic preservation, special effects production of film and television, virtual reality, the removal of redundant objects. At present,the traditional methods of image inpainting have two main categories:the algorithms based on partial differential equations and texture synthesis. However, when inpainting comparatively large areas, the algorithms based on partial differential equations suffer from the blur effect. At the same time, the algorithms based on texture synthesis misses the structural information. This paper, aimed at the disadvantage of the two algorithms refered above, studies several aspects from here after.First of all, by improving the algorithm using the morphological component analysis proposed by M.Elad, we put forward an image inpainting algorithm based on discrete stationary wavelet and wave atoms. The algorithm makes full use of discrete stationary wavelet and wave atoms to represent the structural component and texture component sparsely. Then it solves the corresponding sparse optimization problem to realize the inpainting. The experiment results show that the algorithm can better recover the structural component and the texure component.Secondly, an image inpainting algorithm based on three-layer sparse representation is proposed according to Meyer image model and the theory of multi-scale geometric analysis. According to the Meyer image model, the image can be separated into smooth, edge and texture component. By plenty of experiments, we select discrete stationary wavelet, curvelet and wave atoms to implement sparse representation of these three components respectively. The new algorithm can recover multiple structural components and improve the quality of the inpainted images.At last, this paper improves the idea of non-local means and applies it into the inpainting area. At the same time, combined with the idea of based on pixal, we propose a new algorithm named an adaptive non-local means image inpainting algorithm based on pixal. The algorithm can remove the discontinuity appered in the algorithm using texture synthesis. Moveover, it can make full use of the idea of improved non-local means to synthesis image patches consistently with human perception. Through experiments and simulations, we can see that the algorithm can improve the visual effect and the quality of the inpainted images.