Separable soft shadow mapping

Abstract

We propose an efficient technique for rendering visually plausible real-time soft shadows in screen space. First, we propose a novel blocker estimation technique based on a separable filter. Second, our technique performs a separable Gaussian blur in screen space over the hard shadows produced by the standard shadow mapping technique. Although blurring the hard shadows with a separable filter was done before in the literature using bilateral filtering, we use an alternative approach that minimizes artifacts. Since separated calculation is not possible for all cases, we provide data reutilization criteria based on two user-defined error thresholds called α and β. As a consequence of using separable approaches for both stages of the light visibility estimation, our technique is able to improve rendering performance, especially when high-resolution shadow maps and filtering kernels are used.