Abstract
Amodal completion of occluded angles is explained by a field model of visual interpolation (Gerbino & Fantoni 2000) based on the vectorial combination of good continuation (GC) and minimal path (MP). Here we consider symmetrical and asymmetrical occlusions of 90 deg angles and compare psychophysical data with predictions derived from different 2D-interpolation models, including ours. A multiple-probe procedure was used to estimate the interpolation path as a function of occlusion asymmetry, occlusion proportion, and retinal gap between T-junctions. In every trial observers set a straight line tangent to the amodal contour. Three line orientations were defined by the MP segment and by the bisectors of the two GC-MP angles. In all occlusion conditions, locations of the three tangents were consistent with a smooth interpolation path and incompatible with the misorientation of rectilinear sides that would lead to an interpolation path characterized by a compressed vertex. The interpolation curve was closer to MP when the retinal gap between T-junctions was small. Data were also consistent with the following features of our model: 1) the interpolation path is included in the occluded triangle defined by GC and MP; 2) the intersection between the interpolation path and the median of the occlusion triangle belongs to the tangent parallel to MP; 3) the interpolation path is flattened around the intersection with the median. We compared empirical data for each display to predictions of three models: (a) our field model, that implements the shearing hypothesis for asymmetric occlusions; (b) a local cubic-spline solution; (c) the circular-arc solution by Kellman & Shipley (1991). Asymmetrical occlusion cases confirm predictions derived from the generalized version of the field model of 2-D visual Interpolation proposed by Gerbino & Fantoni (2000).