Item

Abstract

We investigated how varying the number of unique three-dimensional parts within an object influenced recognition across changes in viewpoint. Stimuli were realistically-shaded images of objects composed of five three-dimensional volumes linked end-to-end. Of the five parts within each object, either zero, one, three, or five were qualitatively distinct from other members of the recognition set (e.g., brick versus cone). Non-distinct parts were cylindrical tubes. Independent of the number of distinct parts, the three-dimensional angles between components were different for each object as in Bülthoff and Edelman (1992). In both sequential matching and naming tasks we compared the impact of depth rotations on recognition performance. Separate between-subject conditions were defined based on the number of distinct parts for each member of the recognition set. The No-Parts condition was run on all subjects and served as a baseline for the other conditions. For both tasks, three major results stand out. First, regardless of the number of qualitatively distinct parts there was an effect of viewpoint on recognition performance. Second, the impact of viewpoint change in the One-Part condition was less than that in each of the other conditions. Third, the addition of parts beyond a single unique part produced strong viewpoint-dependent recognition performance that was comparable to that obtained for objects with no distinct parts. Taken together these findings indicate that visual recognition may be accounted for by view-based models in which image-based representations include some qualitatively-defined features.