Approximate methods for improving surface appearance
Peter Shirley, Michael Ashikhmin · 2001
Surface appearance is at the heart of computer graphics. Many methods improve the realism of computer renderings of surfaces. Much effort has been directed towards developing new surface modelling and rendering techniques. Many of these algorithms have been effective, but some of them are also quite complicated. This dissertation presents three methods that improve the appearance of surfaces in computer generated images. These methods lie within the computer graphics domains of reflection functions (BRDFs) and texture mapping. First, we describe a general microfacet-based reflection model which captures the main character of reflection for a wide range of materials. The key contribution that makes this development possible is a new form of shadowing function that describes self-shadowing due to surface microgeometry. We then use these results to derive a simple and efficient Phong-like BRDF, which captures anisotropy and Fresnel effects and contains a non-Lambertian diffuse term. On a different level of surface geometry, we present a novel framework for relighting of the surface microgeometry called steerable illumination textures. By constraining allowed illumination to a linear subspace spanned by a fixed number of lights, it is sufficient to use a simple linear combination of images pre-rendered with these lights. A design of basis lights based on steerable functions is presented which attempts to minimize the number of lights needed without sacrificing image quality.