A novel feature for registration of oblique aerial images under large viewpoint variations

This thesis describes a novel image feature, called 2 Edges and one Corner (2EC), that can be used to establish robust match correspondences between aerial images of urban regions under large projective transformation. As implied in its name, the proposed feature is defined by two line segments and their intersection. The lines are constrained to correspond to the boundaries of building rooftops, while the intersections correspond to the rooftop corners. The specific structure of the proposed feature enables us to utilize both geometrical and image context information surrounded by each feature to robustly establish match correspondences between two or more images. To demonstrate the effectiveness of the proposed feature in the matching problem, test dataset images from Pictometry International Corp. have been utilized. Our experimental results show that an average matching accuracy of over 90% can be achieved. These results are superior comparing to the state of the art, which reports an average matching accuracy of 86.32%.