Convex Grouping Impedes Detection of Stereoscopic Depth

When two image regions are separated by an occluder, the strength of their perceptual grouping behind the occluder depends in part on the possible smoothness of the hidden contour completions (ie, Gestalt ‘good continuation’). We consider if grouping strength also depends on whether the contour completion is convex or concave. We hypothesised that the stronger the grouping between two such regions, the harder it is to resolve their relative stereoscopic depth; and employed accordingly an objective method of relative depth discrimination. The stimulus was in stereo. A horizontal bar in the centre of the image occluded two pairs of planar regions parallel with the image plane. One pair assumed a convex (oval) shape behind the occluder, the other pair a concave (hourglass) shape. The regions in one pair had a slight depth difference. The task was to detect which pair was not coplanar. The convex grouping impeded detection of stereoscopic relative depth (73% vs 86%, F1,10=8.66, p < 0.02). This held even when the convex completion boundaries were less smooth than the concave ones, a result opposite to predictions by Gestalt ‘good continuation’. In a control experiment, the stimulus was viewed with the ‘occluder’ in the background, so grouping was no longer possible. No difference between the two pairs was found. Our results suggest that convexity, known to play a role in figure/ground segmentation, is also significant in perceptual grouping, and can even win out over ‘good continuation’. We also propose an objective method of depth discrimination to study perceptual grouping in general.