Fast Urban Aerial Image Matching Based on Rectangular Building Extraction

This letter presents a new approach for image matching in complex urban environments on the basis of existing rectangular buildings. Matching veryhigh- resolution optical remote sensing imagery is important in urban-related series applications, whereas traditional algorithms don't work well in complex urban scenes. The proposed method conducts image matching by combining the external contour and internal structure to construct regional features that represent a building roof, in which perceptual grouping and unsupervised classification are introduced respectively. A newly proposed Euler number matrix combined with a conventional area index and regional histogram describes the regional features in terms of internal heterogeneous distribution, edge character, and spectral feature. A corresponding similarity measurement is then proposed. Specific experiments in urban areas with repetitive patterns and poor textures prove the effectiveness and robustness of the proposed image matching approach. The proposed technique is applied to image mosaic to confirm its feasibility for further applications.