An Automatic Approach for Building Top Silhouette Extraction Using a Pgvf Snake Model

The recent availability of high-resolution remote sensing technology provides a new data source for urban geospatial data acquisition, which has made it possible to detect buildings quickly and automatically. The building data are used for a variety of applications, including urban planning, land-use monitoring, map updating, and 3D city modeling. However, conventional approaches for building extraction are inefficient due to the high spatial heterogeneity between objects, as well as the increased texture and details from high-resolution images. This paper presents a new, automated method to detect the tops of buildings from high spatial resolution imagery using a Poisson Gradient Vector Flow (PGVF) Snake model. Based on the methodology of overall outline extraction for the tops of buildings from partitioned images, improved Canny edge detection was employed to produce a Boolean edge image, which is the initial condition necessary to calculate the PGVF Snake model for building detection. The methods are illustrated with a QuickBird image over the city of Los Angeles, USA. The results indicated that the proposed approach results in 92.3% completeness and 89.5% correctness compared with 61.5 %and 65.5% for the traditional model and 84.6% and 85.5% for the GVF model, respectively.