Abstract
Purpose: According to Marr (1982), perception of a 3D shape is preceded by forming a 2.5D sketch - a viewer-centered representation involving 3D orientations of surfaces relative to the viewer. The present study examines the relation between perceived 3D shape and a 3D shape reconstructed from perceived orientations of the object's surfaces. Furthermore, it tests a new theory of shape reconstruction based on a 3D sphericality (compactness) constraint. Method: A set of polyhedral objects was used. The polyhedra were characterized by three independent vertices, which refer to the number of degrees of freedom in the problem of reconstructing the shape from a single image (Sugihara, 1986). Subjects were tested in two tasks: (i) reconstructing shapes of three visible faces, and (ii) reconstructing 3D orientations of the faces. If perceived 3D shape is derived from perceived 3D orientation, shape reconstructed in task (i) should match the shape reconstructed in (ii). Results: First, there is a large and systematic difference between shapes reconstructed in tasks (i) and (ii). Second, the perceived 3D shape agrees with the reconstructions produced by a new model, in which a linear combination of sphericality and symmetry of an object is maximized. Sphericality of an object is defined as the ratio of the volume squared to surface area cubed. Conclusion: Reliable percept of the shape of a 3D object is NOT derived from the percept of the object's surfaces.