Item

Abstract

Purpose. Objects that differ by qualitative variations in the configurations of their components are more easily recognized from novel viewpoints than are objects that differ quantitatively. This could result because qualitatively different objects have distinct viewpoint-invariant structural descriptions or because images of these objects contain local diagnostic features that are stable over changes of view. Methods. Subjects studied novel, computer-generated objects composed of four long, thin components connected end-to-end. Associated with each target object was a distractor object that had the same three-dimensional structure but different component shapes in all, three, or two of the four positions. The components in the distractor were either the same as those in the target (but rearranged) or different. Subjects studied six target objects and then performed a 2AFC recognition task in which each target object had to be discriminated from its associated distractor. Both objects were rotated ±40° about the vertical axis from the studied viewpoint. Results. Contrary to predictions of structural-description models of shape representation, the percent of correct responses did not vary significantly with the number of changed positions (63.9%, 64.7%, and 64.7%, for 4, 3, and 2 changed positions) or with the addition of a new component shape (63.2% same components, 65.7% different components). Discrimination was enhanced significantly (79.3%. correct), however, if adjacent, identically shaped components in a target were not adjacent in the distractor. Conclusions. These results suggest that the advantages provided by the variations in component configuration that were studied here were not due to the formation of structural descriptions of the objects. Rather, subjects recognized these objects on the basis of salient local features that were preserved across changes of view.