Abstract
In this paper, we derive a novel framework for jointly estimating the reflectance parameters and diffuse texture of a surface of known geometry and illumination from a single image. Unlike previous work, we do not assume the input of multiple images taken under different single light sources, or input of multiple images taken from different viewpoints, or input of multiple images taken under different states of polarization. Under the assumption of spatial homogeneity of reflectance properties, the algorithm searches for the reflectance parameters that minimize the entropy of the diffuse texture image.
