Surface reflectance classifying under natural illumination

Though a point light source is more suitable to measure the BRDF of the surface, the natural illuminations in the real-world are not point light source and very complex. Fortunately, the complex natural illuminations exhibit some statistical regularity R.O. Dror, et al. (2001). These statistical properties of the natural illuminations lead to predictable image statistics for a surface with given reflectance properties. We develop an algorithm for classifying a surface according to its reflectance from a single photograph under unknown illumination by learning relationships between surface reflectance and certain features computed from the observed image. The statistics of the natural illuminations and the relationships learning are performed in frequency domain because the reflection equation is a rotational convolution and it is convenient to analyze it in space-frequency domain.