Image Matching of Satellite Data Based on Quadrilateral Control Networks

Automatic image matching of satellite data is a difficult process due to the complex characteristics of such image pairs, including significant translation, rotation, scaling and illumination differences. A new method is proposed where, initially, a pre‐registration is performed to coarsely align the input and reference images. Next, a dense set of corner points are extracted using the Harris operator. The strongest conjugate corner points are determined and the accuracy is improved by using least squares matching to construct a precise braced quadrilateral control network in the two images. Once the consistency of the established network pairs has been determined, it is used as a framework to match the remaining corner points. Comprehensive evaluations of the proposed method on a variety of image pairs from different sensors prove its efficiency with sub‐pixel positional accuracy.