Perceptual Categorization Deficit and Disorders of Orientation Processing

The patients in the foregoing chapters have clinically evident problems, and much of what is known about them comes from clinical case descriptions. There are other patients who have been termed "visual agnosic" but whose disorders rarely cause them problems in everyday life. These types of agnosia are demonstrated in experimental tasks, and may be studied in groups of patients delineated by lesion site as well as in individual cases selected for their poor performance on the relevant tests.

In addition to their subtlety, the disorders of this chapter have another feature in common: they are detected using misoriented stimuli and are presumed relevant to orientation constancy in object recognition. Orientation constancy is our ability to perceive an object's shape as constant across changes in the object's orientation relative to us. It is at the heart of the problem of object recognition, since the retinal image with which object recognition begins can change drastically as the object rotates. This is particularly true of rotations in depth (i.e., out of the picture plane) because of foreshortening and occlusion.

Approaches to achieving orientation constancy are of two general types: either transform the perceived image into a format that is orientation-invariant, such that two different views of a shape map to the same orientation-invariant representation, or maintain the image format and apply normalizing transformations to bring the different images of a shape into alignment. In the first category are the use of orientation-invariant features and object-centered representations and in the second is mental rotation (see Farah, 2000, chap. 3, for more discussion of these alternatives). Which approach or approaches are taken by the human visual system is an open question.

5.1 Perceptual Categorization Deficit

De Renzi, Scotti, and Spinnler (1969) first described this disorder, whose main feature is difficulty matching three-dimensional objects across shifts of perspective. In their study, the objects being matched were faces, and the poorest performance on this task was found in right hemisphere-damaged patients with visual field defects, implying that the critical lesions were in the posterior right hemisphere. This initial work was followed by a series of well-known studies by Warrington and her colleagues, who somewhat confusingly refer to the disorder as "apperceptive agnosia." Warrington and Taylor (1973) showed that right posteriorly damaged patients were no worse than normal subjects at naming objects photographed from conventional views like that shown in figure 5.1a, but were, on average, quite poor at naming the same objects photographed from unconventional views like that shown in figure 5.1b. Warrington and Taylor (1978) found that even when patients had recognized the conventional view, they were sometimes unable to see that the corresponding unconventional view was the same object in a matching task. The critical lesion site for this impairment, based on superimposed reconstructions of lesions in Warrington and Taylor's (1973) study, appears to be the right posterior inferior parietal lobe.

In addition to the group studies just described, in which a large number of brain-damaged patients are grouped by hemisphere or quadrant of damage and some summary measure of the performance of these anatomically defined groups is compared with the performance of control subjects, there have also been case studies of perceptual categorization deficit. Warrington and James (1988) present three such cases. Right posteriorly damaged patients performed within normal limits on tests of elementary visual function, including the Efron (1968) rectangle matching task (figure 2.2), but performed extremely poorly on a series of tests of perceptual categorization. They were able to recognize only six to eight out of twenty unconventional views of objects, like the photograph in figure 5.1b, although they named almost all of the same objects (seventeen to twenty out of twenty) when shown them later from conventional perspectives, like the photograph in figure 5.1a. When asked to name silhouettes of objects in an unconventional foreshortened orientation, and when shown the silhouettes of objects being rotated from unconventional to conventional a

Figure 5.1

Usual and unusual views, from the research of Warrington and colleagues (e.g., 1985).

Figure 5.1

Usual and unusual views, from the research of Warrington and colleagues (e.g., 1985).

perspectives, they made many errors and required that the object be rotated to a more conventional perspective before being able to identify it.

5.2 Perceptual Categorization Deficit: Characterizing the Underlying Impairment

On the face of things, perceptual categorization deficit appears to correspond neatly to a loss of orientation constancy. For this reason patients with an apparently selective impairment of this ability were of great interest. Indeed, Warrington's research on perceptual categorization deficit was the only neuropsychological evidence cited by David Marr in his landmark book on vision (1982). He interpreted it as an inability to transform the image representation to an object-centered representation of shape, from which perspective and other aspects of the viewing conditions had been eliminated

There are a number of reasons to question the relevance of perceptual categorization deficit to understanding visual object recognition, and even orientation constancy in particular. Warrington (1985) points out that orientation shifts are not the only manipulations of perceptual quality that pose problems for these patients. She cites unpublished data of Warrington and Ackroyd demonstrating that the matching impairment extends to photographs ofobjects with uneven lighting—for example, the

Figure 5.2

Usual and unusual lighting, from the research of Warrington and colleagues (e.g., 1985).

pair of pictures shown in figure 5.2. Warrington therefore describes the impairment in a fairly general way, as a failure of "perceptual categorization," rather than a failure of orientation constancy, suggesting that patients can no longer categorize perceptually dissimilar images in terms of the distal stimulus object that they have in common. However, even this more general interpretation meets with difficulties, as Warrington herself later noted.

One problem is that these patients are generally not impaired in everyday life. Their deficit is manifest only on specially designed tests. This is not what one would expect if so fundamental an object recognition process were impaired. A second and related problem is that these patients have not, in fact, been demonstrated to have an impairment in matching objects across different views. What, you say? Isn't this the impairment for which this group of patients is known? Although perceptual categorization deficit involves a problem in matching different views of objects, all that has been demonstrated for familiar real objects is a problem match ing a usual to an unusual view. Although one could construct a test in which different usual views of objects must be matched, the tests used so far have always included an unusual view.

The way we recognize all manner of usual object views is swift and automatic; seeing and recognizing feel simultaneous, and metaphors of "visual inference" are just that. In contrast, when people recognize unusual views, the process is protracted, often requiring several seconds, and does sometimes involve conscious inference. Data from control subjects in these studies of unusual view recognition shows a low but nonzero error rate (e.g., Warrington & Taylor, 1973). It is therefore likely that the recognition or matching of unusual views requires a kind of effortful processing above and beyond object perception proper. Such processing might more aptly be called visual problem-solving than visual recognition.

When the quality of visual input is degraded, as it generally is following right parietal lesions, patients may find the problems unsolvable. Support for such an interpretation comes from Mulder, Bouma, and Ansink (1995), who administered a range of perceptual categorization tasks to patients whose visual sensory and attentional abilities had been assessed. They found a strong association between perceptual categorization, on the one hand, and sensory and attentional function on the other. Although two patients with left neglect performed within normal limits on the perceptual categorization tasks, the authors point out that such patients were administered the perceptual categorization tests in a way that minimized the effects of neglect.

In sum, as simple and informative as perceptual categorization deficit first appeared to be, in the end it has not shed much light on visual object recognition. Indeed, following an unfortunate pattern seen many times before in psychology (e.g., the lexical decision task and the P300), the effort to understand perceptual categorization deficit in all its unexpected complexity may have superseded the effort to use it to understand more basic questions about mind and brain.

5.3 Orientation Agnosia

If perceptual categorization deficit is not evidence for a dissociable system of orientation-invariant object vision, are there other neuropsychological syndromes that are? One possible candidate is a disorder that Turnbull and colleagues have named "orientation agnosia," described in a small number of right hemisphere-damaged patients (Turnbull, Laws, & McCarthy, 1995; Turnbull, Carey, & McCarthy, 1997; Turnbull, Beschin, & Della Sala, 1997). Patients with orientation agnosia are able to recognize drawings of objects that have been rotated in the picture plane, but are impaired at recognizing the pictures' orientation. For example, given a drawing of bus, one patient correctly named it but oriented the picture upside-down. The orientation errors are manifest when patients select the orientation for a picture, when they attempt to match pictures, and when they copy pictures. Figure 5.3 shows copies of three geometric patterns made by an orientation agnosic.

The first ofthe two approaches to object constancy described earlier, the use of orientation-invariant shape representations, suggests a tantalizing interpretation of this disorder: it results from preserved orientation-invariant representations, unaccompanied by orientation-dependent representations. This interpretation would explain why patients can recognize the objects (they are achieving orientation constancy by using the first of the two approaches described at the outset), but have lost access to the viewer-centered (or environment-centered) image representations that specify the object's orientation relative to them (or their environment). Turnbull and colleagues are pursuing this account while noting some difficulties with it: If object recognition is being achieved through orientation-invariant features or object-centered representations, then orientation agnosics should also be poor at discriminating the "handedness" ofpictures—for example deciding whether two pictures are identical or are left-right reversed. Yet at least one orientation agnosic performed this task well (Turnbull, Beschin, & Della Salla, 1997). In addition, the orientation errors made by these patients are not random, but show a systematic preference for vertically aligned axes ofelongation, with the flatter, more baselike end on the bottom (Turnbull, Beschin, & Della Salla, 1997). While not inconsistent with the operation of an isolated orientation-invariant system of shape representation, this error pattern is not explained by such a system either.

Finally, we do not yet know whether the preserved object representations in orientation agnosia are invariant over depth rotations. Would patients fail to discriminate the orientation of objects with different sides facing toward them? This would bear on the relevance of orientation agnosia to the spatial invariance of real-world object recognition. Davidoff

Simultagnosia

Figure 5.3

Geometric figures and copies of the figures made by an orientation agnosic studied by Turnbull, Laws, and McCarthy (1995).

Figure 5.3

Geometric figures and copies of the figures made by an orientation agnosic studied by Turnbull, Laws, and McCarthy (1995).

and Warrington (1999) describe a patient who, like those of Turnbull and colleagues, is poor at judging picture orientation and handedness. He is also impaired in the recognition of the kinds of unusual views used to test perceptual categorization, as described in section 5.1. We might expect that preserved orientation-invariant representations, if invariant over changes in orientation along all axes, would permit such a patient to recognize depth-rotated objects. However as noted earlier, the unusual views test orientation invariance across particularly difficult spatial transformations or, when silhouetted, with reduced featural information. A more appropriate test for the present hypothesis would be the discrimination of real or artificial shapes rotated over a range of angular displacements. Further investigation of orientation agnosia will undoubtedly clarify these issues.

5.4 Orientation Constancy and Mental Rotation

The second of the two possible approaches to orientation constancy mentioned at the outset was mental rotation: one's current view of an object is mentally rotated to match a stored image. There is evidence that, at least in some circumstances, people use mental rotation when recognizing misoriented objects. Michael Tarr (Tarr & Pinker, 1989; Tarr, 1995) has pursued the issue with a systematic series of experiments in which normal subjects are taught to recognize novel objects presented at some, but not all, possible orientations during learning. For both two- and three-dimensional objects, subjects' response latencies to identifying the objects is a roughly linear function of the angular distance between the test item's orientation and nearest orientation seen during learning, consistent with a mental rotation process bringing the two representations into alignment. With drawings of real objects that have canonical orientations, subjects also show rotation-like response times as a function of the misorientation of the object, although the function flattens out with repeated exposure to the items (Jolicoeur, 1985). This suggests that both orientation-dependent and orientation-independent representations may play a role in normal object recognition (Jolicoeur, 1990). Recent imaging studies are consistent with this conclusion (Gauthier, Hayward, Tarr, Anderson, Skudlarski, & Gore, 2002; Vanrie, Beatse, Wagemans, Sunaert, & Van Hecke P, 2002), showing both common and distinctive patterns of activation for mental rotation and orientation normalization in object recognition.

One way to evaluate the role of mental rotation in orientation constancy is to investigate the effects of a loss of mental rotation ability on the recognition of misoriented objects. The study of such patients can address the question of whether mental rotation is a normal and essential part of object recognition.

Two case studies are directly relevant to this question, with converging support from a third. One patient, studied by Kate Hammond and myself, had severely impaired mental rotation ability following a large stroke that damaged much of his right parietal, frontal, and temporal lobes (Farah & Hammond, 1988). On three separate tests of mental rotation, he was able to perform the 0 degree control trials, but his performance was barely above chance when the stimuli were rotated as little as 45 degrees.

With real objects his visual recognition was unimpaired, and although his recognition of drawings was mildly impaired, he performed identically with inverted pictures (78 percent in both cases). He also was able to read upside down. He was even able to reinterpret letters according to their orientation—for example, looking at a Z and naming it as a Z in that orientation and also a sideways N. Another patient, described by Morton and Morris (1995), became poor at mental rotation following a left parieto-occipital stroke, yet retained good object recognition ability, including the ability to recognize unusual views.

The reverse dissociation has also been noted: Turnbull and McCarthy (1996) described a patient who had retained mental rotation ability despite impairment in the recognition of misoriented objects. Such a dissociation is inconsistent with the possibility that mental rotation tasks are simply harder tests of the same underlying ability as object constancy tasks, an alternative explanation of the previous two patients' performance. These cases suggest that mental rotation is not essential for object constancy; to the extent that it is used under normal circumstances, it is an ancillary or redundant process.

Was this article helpful?

+1 0

Post a comment