Shadow-based Building Detection and Segmentation in High-resolution Remote Sensing Image

This paper proposes an effective method to extract buildings in high-resolution remote sensing images based on shadow detection. Firstly, a superpixel segmentation algorithm called SLIC is introduced to split the input image into homogeneous patches. LDA-based color features of the patches are extracted for detecting shadow regions. According to the positions of the shadows, an adaptive strategy for seed location and regional growth is developed to accomplish the coarse detection of buildings. Finally, buildings are extracted accurately using Level Set segmentation algorithm. The experimental results prove that the proposed method is applicable in various complicated situations and is more robust and precise as compared with other competing algorithms.