Deshadowing of High Spatial Resolution Imagery Applied to Urban Area Detection

Abstract

Different built-up structures usually lead to large regions covered by shadows, causing partial or total loss of information present in urban environments. In order to mitigate the presence of shadows while improving the urban target discrimination in multispectral images, this paper proposes an automated methodology for both detection and recovery of shadows. First, the image bands are preprocessed in order to highlight their most relevant parts. Secondly, a shadow detection procedure is performed by using morphological filtering so that a shadow mask is obtained. Finally, the reconstruction of shadow-occluded areas is accomplished by an image inpainting strategy. The experimental evaluation of our methodology was carried out in four study areas acquired from a WorldView-2 (WV-2) satellite scene over the urban area of São Paulo city. The experiments have demonstrated a high performance of the proposed shadow detection scheme, with an average overall accuracy up to 92%. Considering the results obtained by our shadow removal strategy, the pre-selected shadows were substantially recovered, as verified by visual inspections. Comparisons involving both VrNIR-BI and VgNIR-BI spectral indices computed from original and shadow-free images also attest the substantial gain in recovering anthropic targets such as streets, roofs and buildings initially damaged by shadows.