Diffuse Skylight as a Surrogate for Shadow Detection in High-Resolution Imagery Acquired Under Clear Sky Conditions

Abstract

An alternative technique for shadow detection and abundance is presented for high spatial resolution imagery acquired under clear sky conditions from airborne/spaceborne sensors. The method, termed Scattering Index (SI), uses Rayleigh scattering principles to create a diffuse skylight vector as a shadow reference. From linear algebra, the proportion of diffuse skylight in each image pixel provides a per pixel measure of shadow extent and abundance. We performed a comparative evaluation against two other methods, first valley detection thresholding (extent) and physics-based unmixing (extent and abundance). Overall accuracy and F-score measures are used to evaluate shadow extent on both Worldview-3 and ADS40 images captured over a common scene. Image subsets are selected to capture objects well documented as shadow detection anomalies, e.g., dark water bodies. Results showed improved accuracies and F-scores for shadow extent and qualitative evaluation of abundance show the method is invariant to scene and sensor characteristics. SI avoids shadow misclassifications by avoiding the use of pixel intensity and the associated limitations of binary thresholding. The method negates the need for complex sun-object-sensor corrections, it is simple to apply, and it is invariant to the exponential increase in scene complexity associated with higher-resolution imagery.

The Depths of Cast Shadow

“…De-shadowing in remote sensing is the recovery of a material's response from within shadow-affected areas of the imagery. Within dark shadow pixels there remains a small signal from the surface material and successful de-shadowing first requires quantification of shadow from within a shadow-affected pixel [5][6][7]. Effective de-shadowing techniques in remote sensing use complex physics-based algorithms that consider sun-object-sensor geometry, BRDF, terrain, and atmospheric correction procedures [8].…”

“…With the advent of increasing spatial resolutions of spaceborne, airborne, and drone sensors, the resultant increase in image detail and complexity can cause exponential increases in the complexity of physics-based de-shadowing approaches. Ideally, de-shadowing techniques that are less reliant upon complex physics-based correction procedures can provide more efficient and less complex methods for removing shadow effects to improve image classification and analysis [7]. Shadow effects are a result of illumination variations so it is a prerequisite to characterise illumination effects before shadow detection and quantification is possible [5,7,9].…”

“…Ideally, de-shadowing techniques that are less reliant upon complex physics-based correction procedures can provide more efficient and less complex methods for removing shadow effects to improve image classification and analysis [7]. Shadow effects are a result of illumination variations so it is a prerequisite to characterise illumination effects before shadow detection and quantification is possible [5,7,9]. Illumination conditions for remotely sensed images can be quantified at the time of image acquisition from the atmospheric conditions at the geographical location of the target and the sun-object-sensor geometry.…”

“…Following the examination of Lynch [13] we assess the colour and depth of cast shadow onto a reference white plate panel with incremental sun-object-sensor angles. Cameron and Kumar [7] use a diffuse skylight vector as a surrogate for shadow depth and this study adopts that methodology. The diffuse skylight is represented as a unit vector and the quantity of diffuse skylight (shadow) in each pixel is determined by the scattering index (SI) that is derived from vector projection using linear algebra.…”

“…In their study, Cameron and Kumar [7] conducted a comparative evaluation of their SI technique against the first valley thresholding method of Nagao [14] and the SMACC (sequential maximum angle convex cone) endmember extraction algorithm of Gruninger et al [15]. The comparison specifically targeted areas of known shadow anomalies within high-resolution ADS40 (50 cm) and Worldview-3 (1.2 m) images.…”