Automatic Building Extraction Based on Shadow Analysis from High Resolution Images in Suburb Areas

This paper describes an automatic method for building extraction from high resolution images. It consists of automatic classification and automatic post-processing using shadow information. Shadow and vegetation are automatically detected; bare land samples are selected manually. In addition, by analyzing the distribution categories in the shifted shadow regions, building sample regions are acquired automatically. Based on these four categories of samples, the classification can be implemented using a SVM classifier, from which the initial building results are extracted. The results are optimized in post-processing, including mathematical morphology processing to enhance the completeness of detected regions, region growth to supplement undetected building regions, as well as non-building removal using shadow rates on the intersection boundaries. Optimization yields the final results. Experimental results indicate thatan approach using shadow analysis can extract building sample regions accurately, completely, and automatically, guaranteeing classification precision. The post-processing strategy effectively improves the completeness of building region detection and removes most non-buildings without shadows. Therefore, the accuracy of the final result increases greatly. In all, the automation of this method is high, and applies to buildings in suburb areas.