Highly Realistic Visualization of Caustics and Rough Surfaces

Abstract

When light is scattered from mirror surfaces, complex optical effects occur, which are found in realistic scenes. They include caustics created by focused reflection, multiple refraction, and high-frequency glare from mirror-like microstructure. However, scenes containing mirror paths that involve chains of interactions with smooth metallic and refractive surfaces pose a significant challenge for visualization because, in this case, light paths are difficult to find. In other words, the probability of sampling an acceptable configuration that includes the camera and the light source is reduced. Caustics and rough surfaces with random patterns of glare due to their mirror microgeometry are visually striking examples of these mirror paths of light. There are many specialized methods capable of visualizing mirror paths of certain types with their own sets of constraints. However, none of the existing methods is suitable for the general case. For instance, photonic maps are a good solution for paths of certain types; however, they can cause undesirable blurring and are not applicable in some important cases, such as caustics on non-scattering surfaces or glare.

This paper presents a combined method for visualization of reflective and refractive caustics, as well as rough surfaces. Stochastic initialization and estimation of sample weight are used to find a solution for complex geometry. Two-pass stochastic initialization is used for rough surfaces with normal maps. At the first pass, the normal map is ignored and mirror paths are found on the original smooth surface. Normal perturbation is randomly selected from the distribution of normals present on the normal map. Gaussian approximation of the entire normal map obtained from the lowest level of the MIP map is used. The second pass, which starts with a rough surface, leads to a correct solution for complex geometry. Thus, the proposed method is extended to chains with rough surfaces and multiple interactions.

As a result, both direct and indirect caustics created by reflection from mirror surfaces are visualized. A more complex caustic scene that uses a non-Lambert reflection model, as well as a combined scene of caustic and rough surface, is also visualized. In the process of visualization, patterns on the rough metal surface are not blurred. The method also enables a very good representation of glossy, rough, and mirror-like materials. The results of visualization of highly realistic scenes are discussed.