Multiplexed Acquisition of Bidirectional Texture Functions for Materials

摘要

The bidirectional texture function (BTF) has proven a valuable model for the representation of complex spatially-varying material reflectance. Its image-based nature, however, makes material BTFs extremely cumbersome to acquire: in order to adequately sample high-frequency details, many thousands of images of a given material as seen and lit from different directions have to be obtained. Additionally, long exposure times are required to account for the wide dynamic range exhibited by the reflectance of many real-world materials. We propose to significantly reduce the required exposure times by using illumination patterns instead of single light sources ("multiplexed illumination"). A BTF can then be produced by solving an appropriate linear system, exploiting the linearity of the superposition of light. Where necessary, we deal with signal-dependent noise by using a simple linear model derived from an existing database of material BTFs as a prior. We demonstrate the feasibility of our method for a number of real-world materials in a camera dome scenario.