Abstract
How do we represent viewpoint changes of a scene? We tested three hypotheses about how people would perform in a test of recognition memory for a specific viewpoint of a scene: 1) the views in memory are stored independently and compared to the test image independently (Independence hypothesis); 2) the views are stored in memory in a viewpoint independent 3-D representation, which is compared to the test image (3D hypothesis); 3) the views in memory are stored independently, but comparison to the test image is a function of the sum of the strengths of the degree of match of the memory views to the test image (Combined hypothesis). To test these hypotheses, two study images were shown from either the same or different viewpoints of a scene. Immediately following, participants indicated whether a test image was identical to a study image. When shown a different viewpoint of the scene, the viewpoint could be taken between the two study images (Interpolation condition) or outside the two study images (Extrapolation condition). In Experiment 1, there were many more errors for interpolated than extrapolated views. In Experiment 2, the two study images were from two different scenes, and were tested with the equivalent different viewpoint conditions. Results confirmed that this effect was not due to a stimulus anomaly or to a lag effect on memory accuracy. We developed memory models of the hypotheses described above and found that the memory performance was consistent with the Combined hypothesis. Therefore, we proposed that the visual system does not construct a common memory representation, however both images are simultaneously involved in the retrieval process.
This work was supported by Grant HD26765 from the National Institute of Health and by a grant from Microsoft Corporation.