Generalized linear perspective shadow map reparameterization

Abstract

Shadow mapping is an efficient approach for real-time shadow rendering, but it usually suffers from inherent aliasing artifacts due to the image-based nature. The aliasing errors of shadow mapping can be classified into two categories: perspective aliasing errors and projection aliasing errors. In this paper, we present the Generalized Linear Perspective Reparameterization (GLPR) of shadow maps to reduce perspective aliasing errors. The generalized perspective reparameterization function is formulized first, in which three parameters are taken into account: 1) the cross angle between the view and light direction, 2) the distance from the projection reference point to the near plane in the perspective warping frustum, and 3) the pixel's depth. The distribution of perspective aliasing errors of all currently existing perspective reparamterizations can be uniformly described by this function. Based on this generalized representation, the direction-dependent warping transform is chosen to keep the linear perspective aliasing distribution when the light direction is not perpendicular with the view direction.