Building extraction from high resolution imagery based on multi-scale object oriented classification and probabilistic Hough transform

Abstract—In this paper, we developed a new building extraction system applied on high resolution remote sensing imagery based on multi-scale object oriented classification and probabilistic Hough transform. This can be divided into two different phases: building roof extraction, and shape reconfiguration. For the first phase, the multispectral and panchromatic high resolution satellite imageries are firstly fused for spatial resolution improvement and color information enhancement. The multi-resolution image segmentation is applied on the fused image, resulting in the formation of the different level of polygon primitives at different space scale, providing different view of the scene at different resolution. In addition to the spectral information, the tone, texture, shape, context information is evaluated in an object oriented manner. The classification is based on a fuzzy rule decision tree classifier. By fuzzy evaluating of the shape, texture, context and spectral information, building roofs are extracted by reconstruction and classification from an appropriate space scale of roof polygon primitives. For the shape reconfiguration phase, we adopt the probabilistic Hough transform to delineate the roof dominant line which shows the major orientation of the specific building roof. According to the dominant line, a building squaring algorithm is applied based on rectilinear fitting of the building boundary. It is shown by our experiment that most rectangular building roofs can be correctly detected, extracted and reconfigured, demonstrating the potential application of the method.