Abstract
Whereas grouping of collinear segments is independent of changes in contrast polarity, results from two paradigms show that the efficient grouping of two coterminating image segments into an L-vertex depends on the image segments being of the same contrast polarity. These effects might be expected from the statistics of natural images in that a change in contrast polarity along a contour is relatively common but rare at the vertex of coterminating segments. Exp. 1: Subjects named line drawings of common objects that were either white or black on a gray background. Gaps were introduced into the contours and contour segments were added to the endpoints to produce either pairs of L or T junctions bounding the gaps. (Ls provide strong evidence for the end of a surface so they should inhibit grouping through their vertex, but not their legs. Matched Ts provide evidence for occlusion—contour can be readily grouped through the stems of Ts.) These added segments were either of the same or opposite contrast polarity as the original lines of the object (so that the segments of the L or T were either of the same or different contrast polarity). When the added segments were of the same contrast polarity, naming performance was better when the gaps were bounded by Ts compared to Ls. However, when the added segments were of the opposite contrast polarity, performance with the Ls improved to that of the Ts. Exp. 2 employed a search task within polygons defined only by outward pointing L-vertices. The Ls, in the target-absent condition, were collinear through their legs with the neighboring Ls. With homogeneous contrast, such vertices readily group into polygons so that the target, an inward pointing L (which does not group with the other Ls), pops out, with no effect of the number of vertices. When the two segments of each vertex were of opposite contrast, search performance declined and became dependent on the number of vertices, indicating a lack of efficient grouping.