Euclidean Distance Transform Soft Shadow Mapping

Abstract

The real-time computation of high-quality soft shadows is a challenging task in computer graphics. The existing approximate solutions for soft shadow computation are prone to aliasing artifacts for small penumbra sizes and banding or light leaking artifacts for large penumbra sizes. In this paper, we aim to minimize these problems with the use of a normalized Euclidean distance transform for soft shadow mapping. We compute anti-aliased hard shadows in the camera view, and, on the basis of the percentage-closer soft shadows framework, we estimate the penumbra size of the shadow silhouette and compute a filtered normalized Euclidean distance transform to generate the soft shadows at the penumbra location. Our approach yields real-time performance and generates soft shadows with less artifacts than related work.