Data-Driven Analysis and Synthesis of Bidirectional Texture Functions
The visual appearance of a materials surface gives a vast amount of information about its properties and enables observers to deduce important characteristics and qualities of the surface. Even subtle changes in the material’s texture and reflectance behavior allow for example to distinguish between surfaces made of real or synthetic leather. Therefore, the reproduction of these effects for digital image synthesis is one of the major challenges in computer graphics.
Towards this end, we will first acquire a large database of surface reflectance characteristics. The main goal of the project is then to derive a statistical model of the space of sampled materials. One practical outcome of this analysis is that the prohibitively large volume of data can be compressed so that it can actually be deployed in graphics applications (without this compression the raw dataset is effectively unusable). However, the more important result of the statistical analysis of the material space will be to develop a dramatically more general representation of materials than is currently available. The goal is to reparameterize the high-dimensional material space to allow perceptually meaningful interpolations between the acquired samples, i.e., to generate new materials that blend qualities of samples from the dataset.