Composite lighting simulations with lighting networks

22 March/April 1998 Recent years have seen the introduction of many different lighting simulation techniques for computing the distribution of light in a virtual environment. For environments without participating media (volumetric effects), simulation requires solving the radiance or rendering equation numerically. Many of the available techniques restrict the lighting effects they simulate by using a simplified version of the radiance equation to speed up the calculation. Examples include radiosity algorithms that only consider diffuse reflections but can achieve fast simulations using advanced hierarchical and clustering methods. On the other hand, Monte-Carlo path or photon tracing techniques simulate illumination by recursive stochastic sampling of illumination in the environment, starting with rays from the virtual camera or the light sources, respectively. Although this technique solves the general form of the radiance equation and can therefore account for any lighting effect, it converges rather slowly and is best used for computing illumination via highly specular surfaces. The basic finite element and Monte-Carlo methods have also been combined in hybrid, two-pass, and multi-pass methods. For example, the irradiance caching technique is based on path tracing but also includes smooth basis functions (similar to radiosity) to interpolate indirect illumination from previously cached computations. Twoand multi-pass techniques try to exploit the advantages of multiple lighting algorithms by combining the effects computed by each. For example, several methods employ a radiosity computation followed by a second view-dependent ray-tracing pass for capturing specular highlights. One method also includes a geometric simplification step for speeding up the radiosity pass. In addition, several techniques include Monte-Carlo path and photon tracing for simulating specular and caustic light paths. Usually, these methods split the description of reflection and/or illumination into separate parts (high and low spatial frequency components), each simulated with a different lighting algorithm before being combined. Note, this split is the same for the complete scene, and only a specific combination of algorithms is allowed. In that sense, the Lighting Networks approach builds on top of these methods. It generalizes the concept of hybrid multi-pass techniques and extends it in several ways. Thus, a Lighting Network does not constitute a new simulation algorithm in itself, it provides a way to best use existing algorithms for a specific environment. During development it turned out that Lighting Networks also provide an ideal environment for developing new lighting simulation algorithms and explicitly showing the similarities and differences between lighting simulation approaches. In the Lighting Networks approach, each lighting algorithm is considered a Lighting Operator or LightOp for short (see the next section for a formal definition). Each LightOp takes illumination information as input and generates new illumination information as output after having simulated part of the global lighting effects in the scene. Thus, many of the lighting algorithms mentioned above can act as a LightOp in our approach with almost no changes. In contrast to traditional monolithic lighting simulations, Lighting Networks permit arbitrary combinations of these basic LightOps in order to obtain a composite lighting simulation. Formalizing the representations in which illumination information can be taken as input and generated as output lets you connect different techniques into a network of simulation algorithms working in combination. This flexible combination easily obtains special lighting effects difficult to compute with a single or a fixed combination of algorithms. In that sense, Lighting Networks generalize the basic ideas of Shade Trees and similar approaches and apply them to a more Lighting Networks combine