Morphological filter based graph cut for building detection

Automatic building detection with satellite and aerial images is a hot topic in remote sensing area. Especially, in recent years generation of high resolution Lidar data increased the success of building detection algorithms. In the proposed approach seed points in high elevation areas are obtained by applying morphological operations on Lidar data. Then, geodesic distances between this seed and other elevation points are calculated. Next, a mask is generated using distance map with shadow and vegetation information. Finally, this mask is given to the graph cut optimization and building detection is performed. Obtained results are compared with state-of-the-art algorithms. Results showed that our algorithm produced comparable precision and recall values with them.