Abstract
Suppose you are shown an outline of a simple 2D shape, and are then asked to estimate where it would balance on your finger. What sort of 3D interpretation does your visual system assign to the shape? In a series of experiments, we discovered that center-of-gravity (COG) estimates for asymmetric 2D shapes are systematically biased, such that long and narrow shape-parts are underweighted, and stouter and rounder shape-parts are overweighted. Why should this be? We theorized that this bias emerges because the visual system assumes that narrower shape-parts in 2D are literally *thinner* in 3D — even when no such depth or thickness information is actually present in the image. This hypothesis was tested in two quite distinct ways. First, the COG-estimation bias was found to depend on the presence of visible contours; compared to their COG estimates for shapes, subjects more accurately estimated the centroids of dot-fields whose convex hulls formed the shapes from the earlier COG-estimation task — even though centroids of dots are formally equivalent to COGs of shapes. Next, we used a 3D printer and laser-cutter to physically build uniformly thick objects from the shapes used in the COG-estimation task, and found that viewing (without handling) these real-life objects during COG estimation moved subjects' estimates closer to the shapes' true COGs. We suggest that the visual system assigns rich, irregular, volumetric interpretations even to simple 2D shapes, and that the nature of such representations can be revealed in surprisingly precise and specific ways by investigating perceptions of balance and stability.
Meeting abstract presented at VSS 2014