View-specific Representations Research Articles

Structural and view-specific representations for the categorization of three-dimensional objects

It has been debated whether object recognition depends on structural or view-specific representations. This issue is revisited here using a paradigm of priming, supervised category learning, and generalization to novel viewpoints. Results show that structural representations can be learned for three-dimensional (3D) objects lacking generalized-cone components (geons). Metric relations between object parts are distinctive features under such conditions. Representations preserving 3D structure are learned provided prior knowledge of object shape and sufficient image input information is available; otherwise view-specific representations are generated. These findings indicate that structural and view-specific representations are related through shifts of representation induced by learning.

The representation of unseen objects in visual neglect: Effects of view and object identity

We provide evidence for long-term priming based on view-specific representations of neglected stimuli. A patient with visual neglect, M.P., was asked to search for a target presented amongst other objects on a table. Subsequently recognition memory was tested for items that were identified and for items missed in search. Items that were missed were rejected more slowly than novel items in the recognition memory task, providing evidence for implicit processing (Experiment 1). Implicit memory for missed items was both item-specific (Experiment 2) and view-specific (Experiment 3), and it was eliminated when there were intervening activities lasting about 1 hour (Experiment 4). There was also an implicit memory for distractors in the search task, which was item- but not view-specific (Experiments 2 and 3) and it lasted for at least an hour, even with other intervening activities (Experiment 4). The data suggest that the representations of neglected stimuli may differ qualitatively from those of nonneglected items, with representations of neglected objects being both view-specific and vulnerable across extended retention intervals. The results support the argument that attention is needed in order to encode object representations that are robust to view transformations and temporal decay or interference.

Distinct Patterns of Viewpoint-Dependent BOLD Activity during Common-Object Recognition and Mental Rotation

A fundamental but unanswered question about the human visual system concerns the way in which misoriented objects are recognized. One hypothesis maintains that representations of incoming stimuli are transformed via parietally based spatial normalization mechanisms (eg mental rotation) to match view-specific representations in long-term memory. Using fMRI, we tested this hypothesis by directly comparing patterns of brain activity evoked during classic mental rotation and misoriented object recognition involving everyday objects. BOLD activity increased systematically with stimulus rotation within the ventral visual stream during object recognition and within the dorsal visual stream during mental rotation. More specifically, viewpoint-dependent activity was significantly greater in the right superior parietal lobule during mental rotation than during object recognition. In contrast, viewpoint-dependent activity was significantly greater in the right fusiform gyrus during object recognition than during mental rotation. In addition to these differences in viewpoint-dependent activity, object recognition and mental rotation produced distinct patterns of brain activity, independent of stimulus rotation: object recognition resulted in greater overall activity within ventral stream visual areas and mental rotation resulted in greater overall activity within dorsal stream visual areas. The present results are inconsistent with the hypothesis that misoriented object recognition is mediated by structures within the parietal lobe that are known to be involved in mental rotation.

When does the visual system use viewpoint-invariant representations during recognition?

One popular model of object recognition [26] claims that the visual system typically describes objects using view-specific representations, but that viewpoint-invariant representations are used when objects can be specified uniquely by the arrangement of parts along a single dimension. In a series of three naming experiments using novel, two-dimensional line drawings, we test this hypothesis against alternative accounts of when viewpoint-invariant representations are used during the recognition of upright and viewplane-rotated objects. Experiments 1 and 2 demonstrate that the number of dimensions along which featural information must be represented is the only stimulus feature that influences the type of representation used, consistent with the Tarr and Pinker model. Experiment 3, however, reveals that the use of viewpoint-invariant representations during recognition is not driven purely by stimulus features, and is at least partly under voluntary control. These data suggest that viewpoint-invariant representations are not automatically invoked by the visual system when the requisite stimulus features are present. Rather, our results suggest that top-down control processes, as well as bottom-up stimulus features, jointly determine the conditions under which the visual system uses viewpoint-invariant representations during visual recognition.

Orientation invariance in naming rotated objects: individual differences and repetition priming.

In naming drawings of complex common objects, unpracticed naming times increase with rotation away from the upright, but this orientation effect is attenuated with practice. In principle, attenuation could result from learning to extract orientation-invariant information or from learning view-specific representations at the trained orientations. We contrasted these approaches by examining repetition priming for prime-target pairs presented on successive trials in either the same orientation (horse at 51 degrees primes horse at 51 degrees) or a different orientation (horse at 154 degrees primes horse at 51 degrees), for two subgroups of subjects. One subgroup showed no orientation effect, even when unpracticed, and a correspondingly high generalization of priming across different views. The other subgroup initially showed high sensitivity to misorientation and little priming across orientations but, with sufficient practice, came to show no orientation effect and complete generalization of priming. Thus, some subjects always used orientation-invariant procedures, whereas others learned to do so.

Rotating objects to recognize them: A case study on the role of viewpoint dependency in the recognition of three-dimensional objects

Successful object recognition is essential for finding food, identifying kin, and avoiding danger, as well as many other adaptive behaviors. To accomplish this feat, the visual system must reconstruct 3-D interpretations from 2-D "snapshots" falling on the retina. Theories of recognition address this process by focusing on the question of how object representations are encoded with respect to viewpoint. Although empirical evidence has been equivocal on this question, a growing body of surprising results, including those obtained in the experiments presented in this case study, indicates that recognition is often viewpoint dependent. Such findings reveal a prominent role for viewpointdependent mechanisms and provide support for themultiple-views approach, in which objects are encoded as a set of view-specific representations that are matched to percepts using normalization procedures.

Object recognition under sequential viewing conditions: evidence for viewpoint-specific recognition procedures.

In many computational approaches to vision it has been emphasised that object recognition involves the encoding of view-independent descriptions prior to matching to a stored object model, thus enabling objects to be identified across different retinal projections. In contrast, neurophysiological studies suggest that image descriptions are matched to less abstract, view-specific representations, resulting in more efficient access to stored object knowledge for objects presented from a view similar to a stored viewpoint. Evidence favouring a primary role for view-specific object descriptions in object recognition is reported. In a series of experiments employing line drawings of familiar objects, the effects of depth rotation upon the efficiency of object recognition were investigated. Subjects were required to identify an object from a sequence of very briefly presented pictures. The results suggested that object recognition is based upon the matching of image descriptions to view-specific stored representations, and that priming effects under sequential viewing conditions are strongly influenced by the visual similarity of different views of objects.