Practical Shadow Mapping

Abstract

In this paper, we present several methods that can greatly improve image quality when using the shadow mapping algorithm. Shadow artifacts introduced by shadow mapping are mainly due to low resolution shadow maps and/or the limited numerical precision used when performing the shadow test. These problems especially arise when the light source's viewing frustum, from which the shadow map is generated, is not adjusted to the actual camera view. We show how a tight-fitting frustum can be computed such that the shadow mapping algorithm concent rates on the visible parts of the scene and takes advantage of nearly the full available precision. Furthermore, we recommend uniformly spaced depth values in contrast to perspectively spaced depths in order to equally sample the scene seen from the light source.