An Object-Based Shadow Detection Method for Building Delineation in High-Resolution Satellite Images

Abstract

In satellite image analysis, detection of shadow plays a critical role in the precise object detection applications. The size and intensity of shadow vary with solar illumination angle and miscellaneous building attributes, which may lead to their misclassification. Hence, there is a need for an accurate and robust shadow detection method. In this paper, an object-based shadow detection method is proposed to extract building shadows and subsequently used to delineate buildings from high-resolution satellite images. In the presented method, a shadow mask is generated at pixel level using a fused ratio map of a visible and a false colour image in HSI colour space. Then, the building shadow map is delineated by an automatic thresholding process on an object level. The results of a qualitative and quantitative analysis reveal the effectiveness and stability of the proposed approach for shadow detection and building delineation in high-resolution satellite images. These satellite images are taken at a different solar illumination angle and for different building attributes. Experimental results show that the proposed method achieves average improvement in F score by 21%, IOU (Intersection over Union) by 16%, and accuracy by 9% as compared to some of the state-of-the-art shadow detection methods.