Abstract
Past research on the perception of surface curvature from binocular disparity and/or motion has focused primarily upon simple surfaces (e.g. quadric surface patches, see Tittle, Norman, Perotti, & Phillips, 1998; Perotti, Todd, Lappin, & Phillips, 1998). In the present experiments, we required observers to judge differences in the local curvedness of separated regions on complex, globally-convex 3-D objects. The shapes of these computer-generated stimuli resembled those of natural objects (i.e., the objects possessed 10–20 qualitatively distinct regions of positive and negative Gaussian curvature) and were optically defined by binocular disparity, motion, shading, and texture. Discrimination thresholds were obtained for interval tasks requiring observers to judge the magnitude of differences in curvedness between local surface regions and for ordinal tasks requiring observers to judge which of two regions possessed the higher curvedness. We also manipulated whether the regions to be judged on any given trial had the same or different local surface shape. The results showed large effects of these manipulations. Performance was poorest in the interval conditions where observers discriminated the magnitude of differences in surface curvedness (Weber fractions for this task ranged from 50 to almost 100 percent). In contrast, the observers' performance for the ordinal conditions was higher (Weber fractions were approx. 30 percent). The highest performance of all (Weber fractions of approx. 20 percent) occurred when the two regions to be compared possessed the same local 3-D shape (both regions were either “bumps”, “saddles”, or “dimples”). Our results demonstrate that human observers do not always possess precise knowledge of local surface curvedness and “good” performance depends critically upon whether metric or ordinal surface structure is judged and upon whether the surface regions to be compared are similar or different in local shape.