A Study On DSM-based Mosaicking Techniques For Remote Sensing Imagery

With the development of aerial and spatial remote sensing technologies, the resolution of remote sensing imagery tends to be more and more high, which brings forward great requirements to image mosaicking. As an important research issue in the automation process of image processing, automatic image mosaicking has become a hot focus in the photogrammetry and remote sensing area. Based on Digital Surface Models (DSM), key techniques for automatic image mosaicking are studied comprehensively in this paper, and the major achievements are as follows.1. Based on HIS radiation adjustment algorithms, a relative radiometric correction method is proposed. Under the precondition of not changing the radiation brightness of images, the method can apply relative radiometric correction to images through the process of altering the RGB order of image pixels, and then rotating the brightness of images in the HIS space. Researches prove that the method can effectively improve the visual consistency of homogeneous images.2. Based on the concept of float-water, an automatic seam-line extraction method is put forward. According to the theory that, under the influence of gravitation, water flow to lower places, rules for automatic seam-line selecting process are designed. Utilizing the position, relative to the center line, of the current node, the method determines the weight for the positions around current node, to provide a basis for the next node. The consistency of buildings is verified by inspecting DSM. In order to avoid such problems as the overlapping between seam-line and building edges and the relief placement of buildings, weighted DSM images are morphologically dilated to ensure the effectiveness of extracted seam-line.3. Based on the directional light selecting theories of polarization filters, a new global image matching method is proposed. Through logic calculation of overlapping binary image pixels, a maximum value, which corresponds to the correct matching position, can be found by this method. The experimental results show that the method is robust and can match the images lacking obvious point features.