Abstract
The disparity fields created by matte surfaces match the surface's true depth profile. However, for specular surfaces the disparities are shifted away from the surface, tracing out virtual ('adanaclastic') surfaces in depth. Previous studies (Blake & Bülthoff, 1990, Nature, 343, 165; Wendt et al, 2008, doi:10.1167/8.1.14) showed that surfaces appear glossier and more realistic if highlight disparities are physically correct. But which specific binocular cues does the visual system use to identify specularities? We computationally analyzed disparity fields generated by irregularly-shaped Lambertian and purely specular objects and found several important differences: 1) the adanaclastic surface is piecewise smooth, but with major discontinuities near inflection points of the physical surface, where disparities go to infinity. Such patches are separated by non-fusible regions which appear aniseikonic. 2) Statistics of vertical disparities (VD) and horizontal disparity gradients (HDG) of specular objects are qualitatively different from Lambertian surfaces. For specular surfaces, the distributions are heavy-tailed, containing large values that often exceed fusibility limits. This suggests specific binocular cues that the visual system could use for distinguishing specular and Lambertian surfaces. In order to test whether these disparity cues affect perceived glossiness, we developed stimuli that allow us to vary disparity fields continuously from perfectly specular (vIPD = 1) to matte (vIPD = 0) and beyond to ‘super glossy’ (vIPD = -1) by varying a single parameter ('virtual IPD'). Crucially, stimuli from the interval (vIPD = [0,-1]) were quantitatively different from true specular materials, but their disparity fields exhibiting the aforementioned discontinuities and extreme VDs and HDGs. We measured gloss discrimination (N = 5) and found that above (vIPD = 0.6) human observers perceived stimuli to be as glossy as physically correct mirror stimuli (vIPD=1). This suggests the brain does not 'know the physics of specular reflection' but instead relies on specific binocular cues.
Meeting abstract presented at VSS 2012