Abstract
T-junction geometry (i.e., the amount of deviation from orthogonality, T-dev) is known to affect the strength of occlusion information (McDermott, Weiss, & Adelson 1998), but remains only a candidate factor for the amodal trajectory's shape. Unlike most visual interpolation models, our field model (Fantoni & Gerbino 2001) can embody T geometry as an effective factor. The field model smoothly interpolates T-stems by chaining resultants of good continuation (GC) and minimal path (MP) vectors. When T-dev is different from 0, two hypothetical mechanisms could be operating:
T-junction normalization. Non-orthogonal stems are shifted towards orthogonality (Hearing 1861). This shift implies a corresponding GC-field's rotation and a consequent modification of the whole set of GC-MP resultants compared to orthogonal T-condition.
Orthogonality-dependent modulation of GC strength. Orthogonal T-stems are better extrapolated than non-orthogonal ones (Gillam 1987). As T-dev increases the GC-field's strength decreases.
To test among these hypotheses we probed interpolation trajectories in 9 displays in which one angle of a diamond was symmetric occluded by another diamond. Support ratio and retinal separation between T-junctions were constant. The 9 displays were obtained by combining 3 occluded angle sizes (70, 90, 110) and 3 T-devs (−10, 0, 10). By definition, T-dev is positive when the surfaces intersection's area is smaller than in the orthogonal case. When T-dev is negative both hypothetical mechanisms predict that the interpolated trajectory is flattened (relative to the orthogonal condition). When T-dev is positive the two mechanisms generate contrasting predictions. Observers judged if a briefly flashed probe was inside/outside the amodally-completed angle.
Independent of the interpolation-angle size, the penetration was smaller when T-dev= −10 than when T-dev= 0, consistent with both mechanisms. Penetration was maximal when T-dev= 10, consistent only with the T-junction normalization mechanism.