Abstract
Pizlo et al. (2014) argue that symmetry plays a critical role in the perception of 3D shape, but there are surprisingly few direct empirical tests of this hypothesis. The present experiment compares the adjustment accuracy for symmetric and asymmetric objects in a shape-matching task. Method: Computer-generated stereoscopic images of 3D polyhedra were viewed binocularly through shutter glasses. On each trial, two objects – adjustable and reference – were presented side by side and the participant adjusted the Z-scaling of the former to match the apparent shape of the latter. The stimulus set included 10 mirror-symmetric polyhedra similar to the stimuli of Li et al. (2011), and 30 asymmetric distortions thereof. The distortions were obviously asymmetric and were produced by randomly displacing in 3D every vertex of the polyhedron. Whereas the quadrilateral faces of the symmetric objects were all planar, those of the asymmetric objects were not, and appeared curved. The stimuli were rendered either against a black background or inside a corridor with textured walls receding stereoscopically in depth. Symmetry and context were manipulated in a 2×2 factorial design in four 90-trial sessions, one per condition. The distance to the reference object was manipulated across trials (0.7m, 1.5m, or 2.3m); the adjustable object was always placed at 1.5m and was physically larger than the reference. Results: All seven observers produced similar qualitative patterns. Although statistical power was high, there were no significant differences between the symmetric and asymmetric conditions and no significant interactions. Viewing distance had a large significant effect – the apparent depths of the reference objects became systematically compressed with increased viewing distance for all stimulus types. Scene context had a significant effect too – the reference objects appeared to have more depth when viewed in the corridor than against a black background, relative to the adjustment object.