Improving Shadow Map Filtering with Statistical Analysis

Abstract

Shadow maps are widely used in real-time applications. Shadow maps cannot be filtered linearly as regular textures, thus undersampling leads to severe aliasing. This problem has been attacked by methods that transform the depth values to allow approximate linear filtering and to approaches based on statistical analysis, which suffer from light bleeding artifacts. In this paper we propose a new statistical filtering method for shadow maps, which approximates the cumulative distribution function (CDF) of depths with a power function. This approximation significantly reduces light bleeding artifacts, maintaining performance and spatial costs. Like existing techniques, the algorithm is easy to implement on the graphics hardware and is fairly scalable.