| Real world surfaces such as tree bark, moss, banana skin, and drying mud have a visual appearance that is often complex due to their fine scale geometric and reflective structure. Changes in viewpoint or illumination, or the passage of time, can lead to further dramatic changes in self-shadowing, specularity and interreflection. Such details, although difficult to model abstractly, can provide a high degree of realism in a synthetic scene. I therefore propose an example-based approach to analyzing, synthesizing, and rendering textured surfaces based on densely sampled image datasets. These are acquired as they vary with illumination, viewpoint, and time. Three matrix decomposition techniques are applied to the bidirectional texture function (BTF) data for compression and comparison, and the idea of a universal basis capable of representing multiple textures is explored. The Time-Varying Texture Function (TTF), is then introduced to address progressive time variation in textured surfaces such as growing grass, decaying banana skins, and oxidizing copper. Algorithms for analyzing TTFs and synthesizing novel TTFs in 2D and on 3D surfaces are presented. Methods for controlling TTFs, allowing surface geometry, environmental factors, or user input to guide synthesis and time scale of change are demonstrated. Finally, the acquisition of physically-based control parameters for TTFs is explored. |
