ABSTRACT
Past investigations into the pigeon's recognition of three-dimensional objects that are pictured at several different views have yielded mixed evidence of viewpoint-invariant performance. In our study, we used single, computer-rendered geons (arch, brick, cone, cylinder, pyramid, and wedge) that involved food reinforcement at only one view (Experiment 1) or at five different views (Experiment 2). One-view training produced reliable stimulus generalization to a majority of nonreinforced views of the geon; five-view training produced reliable stimulus generalization to all nonreinforced views of the geon. These results are discussed in regard to both viewer-centered and object-centered theories of object recognition.
