Abstract
Most research on 3D shape perception has focused primarily on how the brain estimates local shape properties such as depths, surface orientations or curvatures. However, the phenomenology of perceived shape can involve much more than local surface geometry. When we look at a bitten apple, we not only estimate local shape properties, but also segment the shape into distinct parts with different functional meanings (stem, fruit, bite). We can often assign different causal origins to different portions of the shape. For example, we readily interpret the bite as being due to a different generative process than the rest of the apple, as if part of the apple were removed from some ‘original’ or ‘authentic’ shape. Here, we studied the geometrical properties that cause concave regions (‘negative parts’) to be interpreted as ‘bites’ that have removed some portion of an object, rather than intrinsic to the object’s natural form. We used convex, irregular hexagons to measure if and how subjects infer the causal history of unfamiliar 2D shapes. Half of the stimuli were ‘bitten’, by randomly intersecting them with another hexagon and deleting the overlap. We asked subjects to rate on a 10-point scale the extent to which each stimulus appeared to have been bitten. Our data show that subjects can do the task surprisingly well even though the stimuli are ambiguous. We propose several visual cues that subjects might use, such as the ratio between the width and depth of the negative part: the deeper and narrower a negative part the more likely that the shape was perceived as being bitten. We further discuss similarities and distinctions between the detection of bites with occlusion and camouflage. These also cause parts of objects to be missing from the image, but are subject to different physical constraints.
Meeting abstract presented at VSS 2012