Abstract
The relative motion between a moving observer and a fixed, specular surface results in a changing pattern of reflected image texture that appears to aid the interpretation of material properties [Hartung and Kersten, 2002]. Surface reflectance, however, can be judged from static scenes with known geometry using only spatial image statistics [Dror, Adelson, and Willsky, 2001]. To what extent does image motion alone contribute to the perception of surface reflectance? To isolate the contribution of motion from the spatial image statistics we define “specular flow” to be the 2D image motion of a reflected scene. To explore the effect of dense specular flow on the judgment of material properties, we develop novel random-dot stimuli of simple 3D scenes in which monocular image features move according to the 2D specular flow. In natural scenes object surfaces typically exhibit both specular and Lambertian reflectance and thus relative motion between the observer and such a surface results in an optic flow field containing both diffuse and specular components. To model this, we generate random-dot stimuli containing both diffuse and specular flow. Neither set of stimuli are perceived as reflective even when subjects form the correct 3D interpretation of the object and the specular motion. This suggests that spatial image cues are critical for the perception of surface reflectance. To what extent the specular flow of extended spatial image cues contributes to the interpretation of material properties remains an open question.