Automated detection of buildings and roads in urban areas from VHR satellite images

Abstract

In this paper, we present an unsupervised approach to detect regions belonging to buildings and roads in urban areas from very high resolution (VHR) satellite images. The proposed approach consists of three main stages. In the first stage, we extract information that is only related to building regions using shadow evidence and probabilistic fuzzy landscapes. First, the shadow areas cast by building objects are detected, and the directional spatial relationship between buildings and their shadows is modeled with the knowledge of the illumination direction. Thereafter, each shadow region is handled separately and the initial building regions are identified by iterative graph-cuts designed in two-label partitioning. The second stage of the framework automatically classifies the image into four classes: building, shadow, vegetation, and others. In this step, the previously labeled building regions as well as the shadow and vegetation areas are involved in a four-label graph optimization performed on the entire image domain to achieve the unsupervised classification result. The final stage aims to extend this classification to five classes, including the road class. For that purpose, we extract the regions that might belong to road segments and utilize that information in a final graph optimization. This final stage eventually characterizes the regions belonging to buildings and roads. Experiments performed on twelve test images selected from GeoEye-1 VHR datasets show that the presented approach has the ability to extract the regions belonging to buildings and roads in a single graph theory framework.