Abstract
What geometric constraints determine when the visual system will interpolate between two disjoined image fragments to create the percept of a single unified surface? In particular, to what extent does local contour geometry predict surface completion? We compared visual completion (both modal and amodal) across situations in which the local contour geometry was the same, but the surface geometry (i.e., its shape description) was very different. Observers viewed stereoscopic displays consisting of two inducers separated by an orthogonally oriented oval. The oval was given either near or far disparity relative to the inducers, thus requiring the inducers to complete either amodally behind the oval, or modally in front of it. The contours of the inducers leading up to the oval were bent either inwards or outwards, by the same angle, requiring either concave completion (inducing a two-part structure) or a convex completion (with no part boundaries). On each trial, two probes (a single or double “wiggle”) were briefly flashed—one along each inducer—and then masked. Observers judged whether the two probes were the same or different. From single-object-superiority / two-object-cost paradigms, we expect performance to be better for stronger visual completions. Accuracy and RT data revealed that, for both modal and amodal completion, performance was better in the convex case than in the concave case. The results support and extend Liu et al.'s (1999) findings, using a very different method. They indicate that modal and amodal completion depend not only on local contour geometry, but also on the shape description (such as perceived part structure) that the enclosed surface receives.