Abstract: Problems exist in building extraction from high-resolution remote sensing images in terms of spectral features. For example, the same object has different spectra and different objects have the same spectrum, and geometric features of a building may be more typical than its spectral features. Having summarized some typical geometric features of buildings, building extraction is done based on their geometric features with features of area, fitting rectangle, length-to-width ratio, and main directions. The experiment shows that extracting buildings with geometric features is more effective than with spectral features.However, since image segmentation depends on spectral features so that calculation of geometric features is affected by spectral features, building extraction based on geometric features is not satisfactory. As the height feature of a building is the most typical geometryfeatures, introduction of height features into image segmentation and feature extraction is an effective way to improve accuracy of building extraction.
 

Key words: building, geometry feature, spectral feature, height feature, extraction