Abstract
A three-dimensional surface can be described in terms of local depth, orientation and curvature. There has been interest for some time in the relative accuracy of observer judgments for these three attributes. In the present study we compare accuracy for judgments of relative depth, slant and curvature in structure-from-motion displays, using the same basic stimulus and a common psychophysical task for each type of judgment. The stimulus was a vertically-oriented cylinder slanted 20 degrees relative to the horizontal axis, on which dots were randomly positioned. It was shown in parallel projection rotating about a vertical axis. The cylinder was divided into three segments with equal projected areas by two horizontal “gaps.” On every trial either the top or bottom segment was altered in either depth, slant, or curvature. Trials were blocked according to the type of change and observers were informed of the type of change in the block. Observers were instructed to indicate which part—the top or bottom—was different in depth, slant, or curvature from the other two parts. Thresholds were determined separately for detecting changes in depth, slant and curvature. In order to have a consistent metric for comparing thresholds for different surface attributes, we computed the change in the 3-D positions of the dots required to detect a difference in the depth, slant or curvature of one of the parts relative to the other two. The changes in the 3-D positions of the dots required to detect a change was smallest for depth. Larger changes were required for slant and for curvature. This suggests that it is more difficult to detect changes in higher order surface attributes than to detect changes in depth, although it does not necessarily follow that higher order attributes are derived perceptually from differences in depth.
Supported by NIH Grant 1R01EY12437.