Abstract
We investigate how observers assess surface roughness visually. In the past, we have found systematic failures of roughness constancy across changes in scene illumination - more glancing illuminants make surfaces appear rougher (Ho, Landy & Maloney, VSS 2005). Here, we examine roughness constancy across changes in viewpoint in scenes with constant illumination. Methods: The stimuli were 3D surfaces resembling coarse sandpaper composed of small triangular facets. Roughness was defined as the variance of the random facet heights. Stimuli were rendered for binocular viewing both frontoparallel (0°) and from three viewpoints to the right and left (±20, 40 and 60°). A punctate illuminant was located at a fixed position relative to the surface (30° to the left). We used a staircase procedure (two-interval, forced-choice) to estimate the points of subjective equality of perceived roughness for surfaces with various levels of physical roughness across each pair of viewpoints. Results: Across viewpoints on the right-hand side (opposite the illuminant), perceived roughness for a given surface changed little as viewpoint was varied. That is, observers were nearly roughness constant. However, for all other tested comparisons, observers exhibited large, systematic failures of roughness constancy. We found that these failures were well predicted by the changes across viewpoint of (1) proportion of cast shadow, (2) mean luminance and (3) facet shading variance. Our results suggest that, in assessing visual roughness, the visual system mistakenly makes use of “pseudo-cues” to roughness that are not invariant under changes in viewpoint.
Grants EY16165 and EY08266 from the National Institutes of Health