The Line‐Based Transformation Model (LBTM) for image‐to‐image registration of high‐resolution satellite image data

Abstract

Image‐to‐image registration is a prerequisite step for many important applications in different disciplines. In the field of remote sensing, the launch of new high‐resolution satellites raises the need for a fast and dynamic technique for image‐to‐image registration in order to obtain full benefit from these satellites, especially for applications such as Earth monitoring and map updating. This paper presents the utilization of the Line‐Based Transformation Model (LBTM) for image‐to‐image registration following the successful use of the LBTM for the rectification of high‐resolution satellite images using linear features as control features.

The developed model is similar in structure to some other point‐based transformation models. However, line segments on linear features are used first to recover the model transformation coefficients. Based on the recovered coefficients, the whole image is registered using the ordinary point‐based form of the model. Line segments on linear features have been chosen due to their existence in images regardless of the type of the land‐use covered by the images and because line segments can be easily extracted from the images. Moreover, control points may not exist or a complete match between points on images cannot be achieved. Different forms of the developed model are discussed and results using different high‐resolution satellite images from both IKONOS and QuickBird satellites are presented. The experimental results of the new technique show high integrity of the new model and indicate that image‐to‐image registration by LBTM is reliable.

Acknowledgements

This project was fully supported by CERG (Project code: PolyU5167/03E) and project number G‐YX61. The authors are very grateful to Prof. Z. Li, The Hong Kong Polytechnic University; Prof. Clive Fraser, Melbourne University; Dr Hazem Baraket and the Intergraph Division at OpenSoft Company for providing us with the IKONOS and QuickBird images used in this paper. The authors extend their thanks to Dr Bruce King, The Hong Kong Polytechnic University, for his support and the constructive discussion. Finally, the authors would like to express their gratitude to the paper referees for their constructive comments, remarks, and discussion.