Potential of using automatically extracted straight lines in rectifying high-resolution satellite images

Abstract

Rapid advances in remote-sensing technologies have provided the ability to acquire high-quality metric information. This is achieved by forming a relationship between image coordinates and ground coordinates as two-dimensional (2D) and 2D-to-3D transformation models. Most of these models are based on point features. Although point-based transformation models are well developed, it is time-consuming and costly to measure point coordinates manually. However, automatic point extraction algorithms, especially in satellite images, are far inferior in quality. This makes other feature-based transformation models more attractive. Linear features have greater potential than other features not only in automation but also in representation and modelling. In this study we investigated rectifying satellite images with different resolutions and orientations using line-based transformation models (LBTMs). In addition, the relative relationships between linear features were imposed by incorporating geometric constraints. Straight lines were extracted using an automated linear feature extraction algorithm. Experiments were conducted with different sets of straight lines and their results are reported and analysed. Our research reveals that LBTMs provide equivalent results to those obtained using point-based transformation models, especially when geometric constraints are forced. Combined point/line-based transformation models were also investigated and the results are similar to those achieved with either point or line features alone.