Complex shape building extraction with Bayesian structure model

The existing methods for building boundary extraction have limitations in handling non-rectangular complex shape buildings. To address this problem, a method to extract boundaries of complex shape buildings from an aerial image is presented. To account for the irregular structures of complex shape buildings, a Bayesian structure model (BSM) is proposed to represent the relationships of building edges in a probabilistic manner. The Boosting decision tree algorithm is used to fuse multiple types of image features such as color and texture around the building edges, which enhances the robustness of the model. Moreover, a boundary length normalized energy function is designed to describe the building boundary. The boundary extraction is implemented by a global optimization of the function on a graph. The experimental results demonstrate that the proposed method can effectively extract the boundaries of a variety of complex shape buildings.