A Comparison of Techniques for Approximating Full Image-Based Lighting

Light probes, or environment maps, are used extensively in computer graphics for visual effects involving rendering virtual objects into real scenes (Augment Reality). A light probe is a High Dynamic Range omni-directional image covering all directions on a sphere at some location. Each pixel in the light probe image measures the incident radiance at the light probe acquisition point. The figure above shows an example of a light probe image in the longitude-latitude mapping, (similar to an atlas mapping of the Earth). Using the light probe information a virtual object can be rendered with correct scene illumination and inserted into images of the scene with credible shading, reflections and shadows. Rendering virtual objects with light probe information is a very time consuming process. Therefore several techniques exist which attempt to approximate the light probe with a set of directional light sources, the number of which can be chosen to find a suitable compromise between rendering speed and approximation accuracy. This paper evaluates four such approximation techniques in terms of how accurately they perform the approximation as a function of the number of directional light sources. We demonstrate that one of the four techniques is significantly better than the rest when the number of light sources is less than around 50. In general this particular method achieves the same performance as the 3 others with only half the number of light sources