Semiautomatic Building Line Extraction from Ikonos Images Through Monoscopic Line Analysis

This paper proposes a new algorithm for extracting building lines from monoscopic high-resolution satellite images. We focus on extracting lines from rectangular-shaped building roofs with a relatively large size. We achieve this task by devising a unique approach using line voting and matching. Our algorithm works as follows. An input point on a building roof is given manually. A region of interest is defined centered on the input point and within the region lines extracted. Then, a line voting process is applied to estimate initial orientation and position of a building line. The orientation and position are refined by a least squares matching process. We assessed the performance of our algorithm against two Ikonos images. Our algorithm extracted building lines from over 83 percent of buildings tested with average angular accuracy of one or two degrees and average positional accuracy of approximately one pixel. This result supports the algorithm proposed in this paper. The major contribution of this paper is that we provided an alternative way to line grouping approaches when edge responses from buildings are not strong enough.