Study of the dependency of spectral shadow indices on the land cover/use and shadow strength in color aerial imagery

Abstract

Shadow is an important concern in aerial and satellite remotely sensed imagery. Consequently, shadow detection is important for many image processing and analysis operations. Among shadow detection methods, there exist methods that apply a single threshold to a shadow index to determine the shadow cover. Shadowed pixels are affected by shadow and land cover/use, therefore ignoring the land cover/use effect would lead to the wrong analysis of these pixels. The effects of land cover/use and shadow strength on eight color space-based shadow indices have been investigated. Eight shadow indices have been applied to an aerial color image. Subsequently, samples of aerial color images were collected for eight land covers/uses from shadowed and nonshadowed regions, using three shadow-strength classes: strong, medium, and weak shadow. The performance of each shadow index for separating shadowed and nonshadowed samples was examined using a threshold applied separately for each land cover/use and shadow class. All applied shadow indices showed high sensitivity to land cover/use and shadow strength. The results showed that ignoring different shadow strengths (considering the different shadow classes as a single shadow class) leads to some confusion for many color space-based shadow indices and also that a single threshold cannot separate the shadow pixels from the nonshadow pixels. All applied shadow indices show sensitivity to the land cover/use and shadow strength.