| Remote sensing images are very important to the national defense and economic development, and have extensive applications in surveying and mapping, environment monitoring, resources exploring and so on. The processing and application of remote sensing images face a lot of problems that should be solved necessarily. And among them, there are two problems influencing image using. The first one is image information imperfection instauration problems such as image pixel dead or pixel block lost; the other one is to remove the information such as markings, writings, secret objects existing on the image those are redundant or inconvenience for image using. Image inpainting is a technique used for recovering damaged images, removing or camouflaging redundant information. This thesis focuses on the remote sensing image inpainting problem. The main contents include the following:1. Giving a summary of a variety of inpainting problems in remote sensing images, and then carry on some extension; some important theories such as Bayesian principle, spread principle, variation principle, image gradient and difference method involved in image inpainting are introduced, that lay the foundation for my coming research.2. PDE-based image inpainting models including BSCB, TV and CDD are investigated. A new algorithm by combining with convolution-based inpainting and CDD inpainting is proposed. Experiments show that the proposed algorithm meets the results of CDD inpainting and is able to archive faster results than convolution-based inpainting.3. Exemplar-based image inpainting is investigated in detail. And Criminisi inpainting method is optimized. The realization of the optimized algorithm shows a good result for removing important objects in the remote sensing image. And this way present a good way to realize remote sensing image used in both army and civilian. Getting out of the secret of a digital raster graphic is achieved in this method successfully, and also aerial photo sheet number information. |
PDF Full Text Request