Robust building boundary extraction method based on dual-scale feature classification and decision fusion with satellite image

Building edge or boundary extraction is always one of the most important issues for remote sensing application. In order to accurately extract the edges or boundary of a building, there are usually two problems. Firstly, strong interference edges from backgrounds such as road, trees and others cannot be avoided. Secondly, it is more difficult for the lower contrast building edges to be detected. In order to address these two problems to a certain extent, a Robust Building Boundary Extraction method (DS-RBBE) is proposed in this paper, which is based on dual-scale sparse SVM classification and decision fusion. First, training samples are automatically selected by employing prior knowledge of main direction and linearity information. Next, a sparse SVM classifier is trained using the dual-scale local edge features of the training samples. And then, the trained sparse SVM is employed to classify all extracted edges. Finally, a dual-scale decision fusion strategy is performed for final building boundary extraction. In order to evaluate the performance of proposed method, the experiments are conducted on different types of build regions. The results are shown that the proposed method can efficiently extract the building edges and boundaries.