Matching remote sensing images.

Abstract

Image analysis plays a crucial role in many computer vision applications in which images of the same scene with different geometrical orientations need to be compared for further processing. This thesis describes the design and implementation of a model-based vision system for the recognition of aerial images. The main objective is to register two remote sensing images taken at different times. First, some distinctive features are extracted and matched then, these matched features are used as marking points in defining a geometric mapping function. Once registered, the reference image can be used as an aid to automatic interpretation and as a framework for detecting changes between successive images. A two stage matching procedure is used for this task. In the first part, corners are extracted and matched in both images and an initial estimation of the mapping function is computed. This initial function is then used in the second part to estimate the parameters of a global mapping function for the entire image. The process ends when all the extracted features in one image are either mapped to features in the other image, or rejected if no match could be found.