Building extraction by stroke width transform from satellite imagery

Abstract

This paper proposes a novel building extraction method from satellite imagery. Intuitively, the symmetry and regularity of architecture could be used to detect as building area, which benefited from the stroke width transform (SWT) algorithm. Meanwhile, the roof of building are very different with rural area (such as vegetation, wild, etc.) in color space, which can be partitioned by K-means clustering method. The area clustering can obtain the consistency region to complement the discontinuity from SWT algorithm detection. Then, A superpixel generation algorithm is adapted to yield the color distribution of building region, and final building area is able to extract accurately. Different with existing methods, the proposed method performs on a single source satellite image without any other supplement information. Experiment on a large number of satellite imagery demonstrates the efficiency of the proposed method for building extraction.