Stereo-slant: a novel method for measuring figure-ground assignment

Abstract

We present a novel method for measuring figure-ground assignment, based on the observation that the same disparity signal can yield percepts involving different directions of slant depending on how figure-ground is assigned. In figure-ground stimuli containing alternating black and white strips, near disparity was introduced on every other contour (e.g., the left border of each black region). This disparity signal is consistent with the black strips slanted to the right and the white strips slanted to the left. Because adjacent strips share the same border, however, both slant directions are not seen at the same time. Rather the strips of one color are perceived as slanted surfaces, whereas strips of the other color appear as a single frontoparallel surface that extends amodally behind the figural strips. When figure-ground reverses, not only do strips of the other color appear to be in front and slanted, but the perceived direction of slant reverses as well. We manipulated known geometric cues to figure and ground in these displays, including convexity, parallelism, and symmetry (so that one set of strips had one of these properties, but the other did not). Subjects were asked to indicate either which colored regions were slanted, or what the direction of slant was. We found that results with this method nicely tracked these classic figure-ground cues. Our results also suggest that convexity is a substantially stronger cue to figure-ground than either symmetry or parallelism. Our results validate the stereo-slant method as a tool for measuring figure-ground perception, and reveal an interesting interaction between binocular vision and figure-ground perception: regions that are perceived as owning the border also end up "owning" the disparity signal. This can lead to a reversal in the perceived direction of slant when figure and ground switch.

Meeting abstract presented at VSS 2012