Research On Source Image Selection For Remote Image Fusion

Remote sensing is an important way for human to achieve knowledge about the outside world, it involved in many application such as military, weather. But single remote image can't afford enough information to fulfill these applications, image fusion is an important way to solve this problem. Up to now, most researches and studies are focused on the design for image fusion algorithms, the evaluation for image fusion, with many breakthroughes. But when it comes to the source image selection,there are lack of systematic study and theory. This dissertation will devote to this question.Firstly, we study on the image quality assessment, especially on the objective metrics and synthesis methods, for its direct impact on the effectiveness of source image selection. With analysis on its weakness and advantages, we build an image quality assessment system aim at the very issue of source image selection.We aim at 2-source image fusion at first. Based on the information relationship between images, we design two metrics which reflect the information's redundancy and complementarity, and finally achieve a metric reflect the information's ability to replicate the scene integrally, and then design a source image selection approach for 2-source image fusion.Then, aim at multi-source image fusion, which is a more complicated question, we introduce a metric reflect image's ability to replicate the scene prominently. With the integrity metric together, we design a source image selection algorithm. This algorithm select source images from a microcosmic angle, with multi-combination-selection and"survival of the fittest"mechanisms. This approach also study on the end condition.The main contributions of this dissertation are:(1) Systematic and complete analytic and implementation framework are proposed for source image selection. The studies include image quality assessment, metric design, selection framework. We complete the research on image fusion area.(2) We propose two metrics which reflect multi-images'ability to replicate scene, based on integrity and saliency. These metrics achieve meaning on image quality assessment and image fusion effect estimation as well.