Perceptual recognition of familiar objects in different orientations

Abstract

Recent approaches to object recognition have suggested that representations are view-dependent and not object-centred as was previously asserted by Marr (Marr and Nishihara, 1978). The exact nature of these view-centred representations however does not concord across the different theories. Palmer suggested that a single canonical view represents an object in memory (Palmer et al., 1981) whereas other studies have shown that each object may have more than one view-point representation (Tarr and Pinker 1989).A set of experiments were run to determine the nature of the visual representation of rigid, familiar objects in memory that were presented foveally and in peripheral vision. In the initial set of experiments recognition times were measured to a selection of common, elongated objects rotated in increments of 30˚ degrees in the 3 different axes and their combinations. Significant main effects of orientation were found in all experiments. This effect was attributed to the delay in recognising objects when foreshortened. Objects with strong gravitational uprights yielded the same orientation effects as objects without gravitational uprights. Recognition times to objects rotated around the picture plane were found to be independent of orientation. The results were not dependent on practice with the objects. There was no benefit found for shaded objects over silhouetted objects. The findings were highly consistent across the experiments. Four experiments were also carried out which tested the detectability of objects presented foveally among a set of similar objects. The subjects viewed an object picture (target) surrounded by eight search pictures arranged in a circular array. The task was to locate the picture-match of the target object (which was sometimes absent) as fast as possible. All of the objects had prominent elongated axes and were viewed perpendicular to this axis. When the object was present in the search array, it could appear in one of five orientations: in its original orientation, rotated in the picture plane by 30 or 60 , or rotated by 30 or 60 in depth. Highly consistent results were found across the four experiments. It was found that objects rotated in depth by 60 took longer to find and were less likely to be found in the first saccade than all other orientations. These findings were independent of the type of display (i.e. randomly rotated distractors or aligned distractors) and also of the task (matching to a picture or a name of an object). It was concluded that there was no evidence that an abstract 3-dimensional representation was used in searching for an object. The results from these experiments are compatible with the notion of multiple-view representations of objects in memory. There was no evidence found that objects were stored as single, object-centred representations. It was found that representations are initially based on the familiar views of the objects but with practice on other views, those views which hold the maximum information about the object are stored. Novel views of objects are transformed to match these stored views and different candidates for the transformation process are discussed.