抄録
In this paper, we improve the building extraction from scenery images in terms of computational time and energy consumption without decreasing extraction accuracy. The improved method, which is based on the clustering method, also employs GrabCut, which is segmentation algorithm based on graph theory. For acceleration, the image, resolution of which is decreased, is used to drive GrabCut. Our experiment, which employed 106 scenery images, demonstrated that the improved method significantly decreased the computational time by 1.10 s or more compared with the comparative methods. In addition, the energy consumption, which is important to implement the building extraction on devices with limited computational resources, was significantly decreased by 1.18 mWh or more. Furthermore, the improved method did not decrease the extraction accuracy.
