ABSTRACT
This chapter presents a soft shadow algorithm that combines the benefits of these two approaches by employing temporal coherence: the light source is sampled over multiple frames instead of a single frame, creating only a single shadow map with each frame. This approach is faster as typical single sample soft shadow approaches like PCSS, but provides physically accurate results and does not suffer from typical single-sample artifacts. It also works on moving objects by marking them in the shadow map and falling back to a standard single-sample approach in these areas. The main idea of this algorithm is to formulate light-source area sampling in an iterative manner, evaluating only a single shadow map per frame. In order to avoid visible discontinuities when switching from the estimate generated after a disocclusion and the correct result obtained from the shadow buffer Bprev, the two shadow values are blended.
