Euclidean Distance Transform Shadow Mapping

Abstract

The high-quality simulation of the penumbra effect in real-time shadows is a challenging problem in shadow mapping. The existing shadow map filtering techniques are prone to aliasing and light leaking artifacts which decrease the shadow visual quality. In this paper, we aim to minimize both problems with the Euclidean distance transform shadow mapping. To reduce the perspective aliasing artifacts generated by shadow mapping, we revectorize the hard shadow boundaries using the revectorization-based shadow mapping. Then, an exact normalized Euclidean distance transform is computed in the user-defined penumbra region to simulate the penumbra effect. Finally, a mean filter is applied to further suppress skeleton artifacts generated by the distance transform. The results obtained show that our technique runs entirely on the GPU, produces less artifacts than related work, and provides real-time performance.