Recognizing objects seen from novel viewpoints: Effects of view similarity, variability, and time

Seven experiments were conducted to examine (1) the perceptual similarity between different views of novel 3D objects and (2) whether or not the similarity between views is utilized in order to recognize objects seen from novel viewpoints. Experiment 1 demonstrated that the perceptual similarity between different views of an object is not simply a function of the angular disparity separating the views, but rather, a function of the viewing angle (i.e. whether objects face toward or away from a viewer) and the degree of foreshortening among views. Experiment 2 indicated that the recognition of novel views is better accounted for by the similarity between novel and previously experienced views than by the angular disparity between the views. Experiments 3-4.1 showed that experience with more variable views results in improved generalization to novel views. Experiment 5 revealed that recognition advantages due to the combination of previously experienced views does not appear to depend on the views appearing in close temporal succession. Together, the pattern of results suggests that (1) after experiencing one view of an object, subsequent recognition of the object from a novel view is dependent on the perceptual similarity between the novel view and the previously experienced view and (2) generalization to a novel view is improved by combining previously formed representations of more variable sets of views, regardless of the amount of time separating these views.