Abstract
Purpose. We investigated the way in which an object representation is encoded in order to allow recognition when the same object is later encountered from a different viewpoint. In particular, we investigated recognition when a canonical-view of the object was first seen, followed by a noncanonical-view of the same object; and vice versa. We also manipulated similarity between objects to study the degree to which this generalization was constrained by shape details. Finally, we manipulated experimental instructions to discourage view generalization, in order to assess the extent to which the view generalization was involuntary. Method. In Experiment-1, subjects rated the goodness-of-view resulting in our definition for canonical- and noncanonical-views for the subsequent experiments. Basic-level and subordinate-level objects were run in separate blocks for Experiment-2 (with viewpoint-independent object-based-recognition task) and Experiment-3 (with viewpoint-dependent image-based-recognition task). Each block had a study-phase (attractiveness-rating task) and a test-phase (recognition task). In study-phase half of the objects were shown in canonical-view and half in noncanonical-view. In test-phase old objects and equal number of novel objects were shown. For old-objects half images were in the same viewpoint as study-phase and the rest were in the other view. Results. In the object-based task, for basic-level objects, view generalization was perfect between canonical and noncanonical views. However, for subordinate-level objects, the generalization was limited. It was also significantly better when the objects were first seen in a noncanonical-view and tested in a canonical-view, than the other way around. In comparison, in the image-based task and for basic-level objects, there was more involuntary generalization from noncanonical to canonical views. For subordinate-level objects, this generalization was comparable between canonical and noncanonical views. Conclusions. Results imply that in shape-representation, a view was less likely to be stored as a snapshot-template, but more likely to be involuntarily “canonicalized” in viewpoint-generalization.