Abstract
When viewing a specular object, such as a highly polished kettle, the retinal image consists of a distorted version of the surrounding environment. If viewing binocularly, a given environmental feature is generally reflected to the left and right eyes from different surface locations, leading to a pattern of disparities that does not coincide with the true physical location of the surface. Previous psychophysical work (Blake & Bülthoff, 1990, Nature, 343, 165) suggests that specular reflections can be used to constrain the 3D shape interpretation of partially-specular surfaces. However, here we ask how observers perceive 3D shape when viewing wholly-specular objects that give rise to complex patterns of disparity. In particular, we ask whether participants (a) recover the physical location of the surface or (b) perceive the surface location at the disparity-defined distance. First, we report the results of mathematical simulations designed to quantify the disparity fields produced by viewing smooth 3D objects (“potatoes”) binocularly. We develop a method for quantifying binocular disparities based on matching the reflected ray vectors for the two eyes, illustrate how specular surfaces give rise to complex, and often abrupt, changes in disparity, and discuss the matching ambiguities that are inherent when viewing specular objects. Second, we report the results of psychophysical experiments in which observers (n = 6) used a disparity-probe method to indicate their perception of specular 3D shapes (n = 9). Observers made multiple settings for a hexagonal grid of points (n = 61) that overlaid the objects. We find that observers neither (a) recover the physical surface location nor (b) match disparities. Our mathematical simulations suggest that disparity signals are differentially reliable at different locations on an object's surface, suggesting that observers' settings may result from the propagation of depth estimates from reliable portions of the object.