Automatic registration of airborne images with complex local distortion

Accurate registration of airborne images is challenging because complex local geometric distortions are often involved in image acquisition. In this paper, we propose a solution to this registration problem in two parts. First, we present an area-based method to extract sufficient numbers of well-located control points, and second, we use the extracted control points with local transformation models to register multi-temporal airborne images. The proposed image registration methods were applied to two airborne images with complex local distortion. Performance was evaluated and compared using different transformation models (global models and local models), different numbers of control points, and different similarity measures (correlation coefficient and mutual information). The results showed that local models outperformed global models, more control points could significantly improve local transformation models but not on the global transformation models, and two similarity measures performed similarly. These results revealed two important findings: first, the area-based methods generated larger amounts of evenly distributed control points; and second, local transformation models achieved better registration accuracy when larger amount of evenly distributed control points are used. We concluded that the combination of area-based control point extraction with local transformation models is effective for the registration of airborne images with complex local distortion.