Improving Image-Space Caustics Via Variable-Sized Splatting

Interactivity requires tradeoffs to achieve the right balance between rendering quality and speed. In practice, today’s applications restrict lighting to mainly direct illumination, sometimes augmented by precomputed transfer techniques for diffuse global effects. Dynamic high-frequency specular effects, such as caustics, are largely lacking due to the high costs for recomputation each frame. Recent work has introduced a variety of related caustics approximations that interactively render light-space photons into a photon buffer, gather them into a caustic map, and project this map onto the scene similar to shadow mapping. While the process is simple and straightforward, the discretization of light into a finite number of uniformly-distributed photons leads to undersampling and aliasing artifacts. This paper examines two techniques for reducing these artifacts using varying sized photon splats. Conceptually, these are similar to the variable-radius k-nearest neighbor search used in photon mapping, allowing noise reduction in areas of low photon density while maintaining crisp caustics at focal points. Our techniques improve image quality at a modest cost that is significantly cheaper than supersampling the photon buffer.