Abstract
Determining whether a material is matte or shiny is theoretically under-constrained. Image intensity variations could be due to paint changes across a uniform matte surface or illumination variations reflected in a uniform specular surface. Remarkably, the human visual system easily distinguishes shiny from matte objects. How does it do this? We used measured natural illumination maps from Debevec's Light Probe Image Gallery[1] to investigate three sources of information for seeing an object as shiny: 1) Consistency between the background environment and the reflection; 2) “Naturalness” of the illumination environment; 3) Optic flow.
Because shiny objects reflect their environment, the pattern of colors across the object and the illuminating environment are correlated. We show that human vision often doesn't care: e.g. an object reflecting a museum interior appears to be shiny even when shown against forest, or other inconsistent backgrounds.
The sufficiency of internal region pattern for indicating shininess raises the question of what class of illumination maps are best for perceiving shininess? Dror et al.(2001)[2] have shown that natural illumination maps have characteristic non-gaussian, non-stationary statistical properties. We show that departures from “naturalness” can have striking effects on perceived shininess. For example, a shiny object in a “white noise” illumination world looks matte.
A rotating shiny object projects different optic flow patterns than a rotating matte object. We “painted” objects with illumination maps such that for any given static view, the object appeared shiny. However, when the apparently shiny object begins to rotate, it immediately appears matte, and when it stops, appears shiny again. The visual system sees material from motion.
[1] http://www.debevec.org/Probes/
[2]
DrorR. O.AdelsonE. H.WillskyA. S.Estimating surface reflectance properties from images under unknown illumination. In SPIE Conference on Human Vision and Electronic Imaging, San Jose, CA, 2001.
Supported by: NIH RO1-EY12