Abstract
Previously, we found that observers, when viewing movies of rotating specular superellipsoids of varying corner-roundedness, perceived more-rounded shapes as less shiny than cuboidal shapes. Here we investigate which properties of the stimuli predict observers' percepts.
Our analysis is based on the observation that rotating specular cuboidal and ellipsoidal shapes give rise to image velocity distributions that differ in their characteristics. Image sequences of objects that are perceived shiny tend to have bimodal velocity distributions, with one of the velocity components (“slip”) opposite to motion of surface points on the object, whereas matte-perceived objects have velocity distributions dominated by the motion of the surface points.
To quantify this observation we analyzed the motion information in the stimulus image sequences using steerable spatio-temporal filters. Using linear discriminant analysis, we show that filter responses can be used to correctly classify stimuli as shiny or matte, and show good agreement between these classification results and the observers' ratings of shininess.
To better understand the origins of this motion information, we investigated the relationship between specular flow and shape analytically. We found 1) that specularities appear to flow along the directions that the surface's curvature gradients moved during object rotation. 2) Consistent with our qualitative observations, we find that the image velocity distribution of a specular feature on a rotating shiny object transitions from unimodal to bimodal as the shape of the object varies from ellipsoidal to cuboidal, which may account for the misperception of material in ellipsoidal objects.
Taken together our findings provide an account of how material percepts can arise from image velocity patterns, and what physical properties of objects give rise to these patterns.
This work was supported by NIH grant EY015261. Partial support has been provided by the Center for Cognitive Sciences, University of Minnesota.