Academic journal article Memory & Cognition

Learning Scenes from Multiple Views: Novel Views Can Be Recognized More Efficiently Than Learned Views

Academic journal article Memory & Cognition

Learning Scenes from Multiple Views: Novel Views Can Be Recognized More Efficiently Than Learned Views

Article excerpt

In two experiments, participants were trained to recognize a playground scene from four vantage points and were subsequently asked to recognize the playground from a novel perspective between the four learned viewing perspectives, as well as from the trained perspectives. In both experiments, people recognized the novel view more efficiently than those that they had recently used in order to learn the scene. Additionally, in Experiment 2, participants who viewed a novel stimulus on their very first test trial correctly recognized it more quickly (and also tended to recognize it more accurately) than did participants whose first test trial was a familiar view of the scene. These findings call into question the idea that scenes are recognized by comparing them with single previous experiences, and support a growing body of literature on the existence of psychological mechanisms that combine spatial information from multiple views of a scene.

People are able to recognize previously learned places from perspectives that they have not experienced. For example, after having approached a favorite picnic spot in a park several times from the north and from the east, a person will likely recognize this spot as he or she approaches it from the northeast. The psychological processes that enable this sort of place recognition are currently not completely understood, although two classes of models have been proposed. By one account (which we will call a normalization approach), people store a relatively large number of specific examples of their experiences. Recognition from a novel perspective can then occur by matching the current view to a particular stored view (see, e.g., Diwadkar & McNamara, 1997; Tarr & Pinker, 1989). Empirically, normalization processes are indicated when recognition performance declines monotonically with the distance between a given novel view and its nearest learned view. Another class of models that represents a view combination (or view interpolation) approach holds that people do not rely on single instances of their prior experience; rather, novel views of a scene activate multiple stored views (e.g., Bülthoff & Edelman, 1992; Edelman, 1999; Hintzman, 1986; Ullman, 1998). By this approach, recognition is based on the summed activation (modeled proportionately to similarity) of the novel view to the stored views. As a result, some novel views can be recognized at least as well as familiar views.

Friedman and Waller (2008) provided initial evidence for view combination in scene recognition by exposing participants to views of a playground scene that had been taken from two ground-level perspectives (e.g., 48° apart). Subsequent recognition of these trained views during a test phase was not statistically different from that of novel views of the playground taken from an interpolated viewpoint-one that was between the two trained perspectives. However, novel extrapolated views-those that were outside of the training range-were subsequently recognized less efficiently than were the trained views. Because interpolated views were recognized more efficiently than extrapolated views, yet both were equidistant from the training views, Friedman and Waller concluded that their findings were not consistent with a normalization account of scene recognition. Instead, they concluded that the recognition of coherent real-world scenes was aided by view combination mechanisms that integrated information from the separate trained views.

In the present article, we provide strong additional evidence for view combination mechanisms in human spatial memory by showing that under appropriate circumstances, novel views of a scene can be recognized even more efficiently than can views that have been previously seen and learned. Such a finding is not predicted by normalization accounts of scene recognition; however, it is possible, in principle, according to most models of view combination (Edelman, 1999; Edelman & Bülthoff, 1992). …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.