Building Extraction from Very High Spatial Resolution Image

The new availability of very high spatial resolution satellite images offers a mapping potential for scales reaching from 1: 5000 to 1: 10000. The urban man-made objects such as buildings can be delimited. However problems and difficulties can appear, particularly in the high local variance environment and spectral signatures disturbances context. The extraction methods should be adapted to these new images. Two principal techniques are currently explored for automatic building extraction from very high spatial resolution satellite images: the “zonal” approach and the improvement of per-pixel classification. This paper proposes an original detection approach of building’s centres based on variance features. A unique parameter is used, taking into account jointly the variance of building and its close neighbourhood. The proposed method has been applied to a panchromatic IKONOS image (1 metre resolution) in an urban area. Although the methodology is not entirely completed and needs additional developments, the preliminary results encourage us to continue the research in this direction.