Building Extraction From RGB VHR Images Using Shifted Shadow Algorithm

Building extraction from RGB VHR images is an important and popular topic for mapping, disaster emergency responding, and city management. The automation of most methodologies cannot meet the need for applications. In this paper, based on classification and optimization, we propose a novel methodology using shadows to automatically extract building samples and verify buildings accurately to improve automation and accuracy. On one hand, in order to acquire various and reliable building samples automatically for classification, detected shadows first are shifted opposite to the direction of illumination to extract building shadows. Furthermore, each building shadow will be shifted again in the same way. Then according to the distribution of classes in these customized shifted regions, building samples can be filtered out by removing those recognized objects. On the other hand, besides the common measures to optimize the initial building during post-processing; a new, original, and an efficient shadow-based index for building verification is also designed. Shadow rate on the intersect boundary between the expanding edge of candidate regions and their shifted regions following the illumination direction can efficiently recognize buildings. When the proposed method is compared to other sample acquisition methods based on shadow, experimental results show that the approach for building samples acquisition is helpful to get accurate initial building results. Moreover, in comparison with other building extraction methods, the proposed building verification method can distinguish buildings from non-buildings. This significantly improves the accuracy of the final results. Numerical assessments performed on a series of test images indicate that our proposed approach for building extraction is efficient and feasible, especially in suburban areas.