Research Of Image Inpainting Algorithm Based On Analysis Of The Local Characteristic And Sparse Transformation

The origin of image inpainting traces back to 1980s',which was used to repair damaged artworks and paintings. With the rapid development of computer technology, the world has been digitized increasingly. Synchronously, Digital image inpainting technology has come to the front gradually. Today, digital image inpainting technology has become an important branch in the field of digital image processing,which is widely used in various fields, including restoration of the damaged image, removal of a certain object, image coding and compressing, image super-resolution analysis and so on.Firstly, the research background and significance of digital image inpainting technology are introduced in this paper, and then the development course of the technology is briefly reviewed too. Nowadays, the algorithms of digital image inpainting are various, in which there are two most important kinds.One is the image inpaiting based on partial differential equations (PDE), the other is the image inpainting based on texture synthesis. Several classic inpaiting algorithm model is given in chapter 2 of this paper.The image inpaiting based on exemplar patch is an important branch of the image inpaiting based on texture synthesis. Chapter 3 describes the algorithm in detail, and then gives some improvements about the realization process. The experimental results show that the improved algorithm has strong capability to keep strong edge of damaged image. The image inpaiting based on partial differential equations is suitable to inpaint the small scale regions in images. However, this kind of algorithm is often based on mathematical theory models which are difficult to be understood. According to this issue, a simple algorithm of image inpaiting based on information diffusion along the isophotes direction is proposed by analysing the local characteristics of image in chapter 4, and achieves good results.It is a novel approach to inpaint damaged image using sparseness in some thansform domain. In the last part of the paper, an iterative algorithm based on sparse transformation is discussed. According to the problems existing in the experimental results, an iterative algorithm of image inpaiting based on hyper-wavelet transform and isophotes smoothing is proposed. The experimental results show that the algorithm works efficiently on small scale regions.