SECOND GENERATION CURVELET TRANSFORM FOR BUILDING EXTRACTION FROM HIGH RESOLUTION SATELLITE IMAGERY

摘要

The process of automatic extraction of buildings from digital imagery has had major practical importance for many years in the area of data acquisition and updating of geographic information system (GIS) databases. This process also presents a huge scientific challenge for researchers as a result of the heterogeneous nature of the buildings, especially in the developing countries[Aytekin et al. 2009].Several techniques were used in building extraction for satellite images in this research paper, second generation curvelet transform will be introduced as an edge detection tool helping in the detection of buildings boundaries. Second generation curvelet transform provides optimally sparse representations of objects, which display smoothness except for discontinuity along the curve with bounded curvature [Candes et al. 2006]. Some papers have investigated this technique for edge detection in high resolution satellite imagery such as IKONOS or QuickBird, and microscopic imagery [Geback and Koumoutsakos 2009; Guha and Wu 2010; Zhenghai and Jianxiong 2009; Xiao et al. 2008], which show great potential of using curvelet transform in solving edge detection problems. However; until now, there is no research tackling the building detection problem using the curvelet transform in high resolution satellite imagery.The algorithm consists of four main parts; first, data fusion between the panchromatic band (0.50 m resolution) and the multispectral ones (2.00 m resolution), to generate 8-spectral bands with a resolution of 0.50 m. Second, a Gaussian high pass filter is applied to enhance the edges. Third, using the curvelet transform, edges will be detected based on the fact that the values of curvelet coefficients are determined by how they are aligned in the real image. The more accurately a curvelet is aligned with a given curve in an image, the higher its coefficient value. Fourth, a filling process is performed for every closed boundary followed by calculation of statistics for these enclosed boundaries; such as area, major and minor axis, and compactness to extract the buildings.