hide
Free keywords:
-
Abstract:
Previous studies suggest that there are "canonical" viewpoints from which objects are identified most readily. Object naming has been the primary objective measure used to assess view canonicality, but this task has not proven adequate in distinguishing the many explanations of this phenomenon that have been offered. We examine object naming (Experiment 1a) and name verification (Experiments 1b and 2) to better understand the origin and nature of canonical view effects in recognition. In the name verification
experiments, observers read an object name and then viewed an image of an object and decided as quickly as possible whether the image matched the name. The stimuli were images of 3D computer models of seven common objects. Each object was rendered from one canonical viewpoint2 (determined in
a separate experiment by Blanz, Tarr, Bülthoff, Vetter, 1996) and two noncanonical viewpoints. Observers named the objects faster in canonical views, but performance was not affected by viewpoint in either name verification experiment, even on the first presentation of each view. We
interpret these results in terms of a view-based similarity model. Naming is slow for noncanonical views because they are similar to stored views of more than one object, leading to response competition. The name verification task
reduces the space of relevant views in long-term memory that must be compared to an input view, which for the views studied here minimized the likelihood of confusions and eliminated differences in response times for different views.
