Abstract
Many objects in our environment undergo visible processes of change, such as growth, weathering, bending or crumpling. Despite huge resulting changes in appearance, we can often identify both the object across the transformation and the transformation that has been applied to it, as if the visual system decomposes the object’s features into distinct causal contributions. For example, a concavity in a cookie is readily perceived as a ‘bite’ whereas a similar concavity in a croissant is interpreted as a ‘bend’. If—as we suggest—the visual system infers generative models to ‘understand’ shapes and their origins, different interpretations of parts of a shape (e.g. a concavity) should have profound effects on its perceptual organization. Here, we test the effects of causal interpretations on perceived symmetry. In one experiment, subjects viewed familiar symmetric 2D shapes, with a probe dot placed at different locations on the outline on every trial. Some objects appeared complete while others had a portion removed (‘bitten’). Subjects had to place a second dot to indicate which location appeared symmetric to the probe relative to the shape’s global symmetry axis. In a second experiment, using irregular shapes, subjects directly reported each object’s principal symmetry axis by distributing a series of dots along the axis. We compared responses between pairs of objects that had small differences in shape but large reported differences in their perceived causal history: each pair contained 'bitten’ and 'non bitten' versions. In both studies we find that concavities that are interpreted as 'bites' are suppressed from symmetry axis judgments. In other words, the visual system compensates for ‘missing portions’ of both familiar and unfamiliar objects. In contrast, geometrically similar concavities that are interpreted as intrinsic to the object are incorporated into symmetry axis judgments. This suggests that inferred generative processes substantially influence perceptual organization.
Meeting abstract presented at VSS 2015