Real-Time Soft Shadows with Adaptive Light Source Sampling

Simulating physically accurate soft shadows in Computer Graphics applications is usually done by taking multiple samples from all over the area light source and accumulating them. Due to the unpredictability of the size of the penumbra regions, the sampling density has to be quite high in order to guarantee smooth shadow transitions in all cases, making the process computationally extremely expensive and therefore hardly usable in anything else than offline rendering aplications. Thus, we suggest a new approach, in which we select the sampling points adaptively and avoid the generation and evaluation of shadow maps which do not contribute to an increased image quality. The main idea is to reproject the shadow maps of two neighboring sampling points into camera space, investigate how much they differ by exploiting the functionality of hardware occlusion queries, and recursively add more sampling points only if necessary. While sampling the light source with a fixed number of samples leads to either over(guarantees constant quality) or undersampling (guarantees constant frame rates) in most cases, our method is capable of selecting only the samples which really contribute to an improved shadow quality. This saves rendering time, but still generates shadows of comparable quality and accuracy. Even though additional calculation time is needed for the comparison step, interactive frame rates are possible in most cases, since the calculation of many unnecessary samples is saved.