An evaluation of local features on satellite images

Satellite image matching is an important task when integrating information from different sensors, viewpoints or taken at different times. This paper presents an evaluation of local features on satellite images. Different blob, edge and corner based detectors have been tested against many variables and parameters. During the extraction process, the detected keypoints are sub-sampled in order to solve the high hardware requirements and the processing time problems with high resolution satellite data. The matching process showed variant performance for the detectors where SURF, FAST, GFTT and Harris detectors have recorded the best matching performance on optical images in terms of higher number of inliers, less running time and robustness to noise. On the other hand, SIFT detector has recorded better performance on radar images. The Experiments were done on different optical and radar images from Rapid-Eye, Pléiades, and TerraSAR-X satellite data for the area of Berlin Brandenburg airport.