Improved anti-aliasing for Euclidean distance transform shadow mapping

Abstract

High-quality, real-time penumbra rendering remains a challenging problem in computer graphics. Existing techniques for real-time fixed-size penumbra simulation generate aliasing, banding or leaking artifacts that diminish the realism of shadow rendering. Euclidean distance transform shadow mapping aims to solve that by using a normalized Euclidean distance transform to simulate penumbra on the basis of anti-aliased hard shadows generated by revectorization-based shadow mapping. Despite the high visual quality obtained with such a technique, the anti-aliasing provided by shadow revectorization comes at the cost of shadow overestimation artifacts that are introduced in the scene. In this paper, we propose an improved algorithm for Euclidean distance transform shadow mapping by reformulating the visibility function of revectorization-based shadow mapping. Through an additional detailed analysis of the results, we show that we are able to reduce shadow overestimation artifacts for penumbra simulation, generating shadows with higher visual quality than previous fixed-size penumbra shadowing methods, while keeping real-time performance for shadow rendering.