Building Detection from Satellite Images based on Curvature Scale Space Method

Abstract

Remote sensing, which provides satellite and aerial images, has recently become the subject of intensive scientific research. It provides an important source of information for the geographical and topographical description of territories, and it allows the collection of information to be used to understand the changes in the environment. This paper deals with the segmentation of satellite images into various textures (building, road, grass, forest, etc.). The automatic extraction of buildings from satellite imagery in urban and suburban areas is the main objective of this paper; it presents a novel approach for building detection using corner detection after the reduction of shadow. The first step is the pre-processing of satellite images by using the Gaussian filter so as to reduce the effects and noises due to the atmospheric components on the electromagnetic radiation, then to increase the contrast between contours to detect shadows, which are the main obstacles in recognizing buildings, by using adaptive thresholding on HSV space color in order to get only the shadow regions; in the second step, points of interest of buildings are detected by using a modified Curvature Scale Space (CSS) for corner detection with a contours-based method. The proposed method is tested with images captured from Google Earth and from some published test images. Some other corner detection methods are used in comparison with the results of the proposed method, and the results indicate that the proposed method works robustly and efficiently.