Figure 411

Kosslyn et al.'s theory of high-level vision. Adapted from Kosslyn et al. (1990).

from visual disorders. It is also an advance on the earlier model in that it provides a detailed process model which can be applied to object recognition, object naming, and object categorisation.

There is another key advantage of the Humphreys et al. (1995) model. It is better equipped to handle the existence of patients with intact functional or semantic information for objects when presented with their names but greatly impaired visual information about objects (see earlier). In the Humphreys et al. (1995) model, visual or structural information is stored separately from functional (or semantic) information, and functional or semantic information is closer than visual or structural information to name information. Thus, it is entirely possible for object names to activate functional but not visual information.

Humphreys et al. (1995) found that "lesions" to the connections between structural descriptions and semantic representations in their model produced a pattern in which access to visual information when an object was presented was essentially intact but there was impaired access to semantic information. This corresponds to the pattern found in some studies on patients with optic aphasia (e.g., Hillis & Caramazza, 1995). For example, such patients are good at at the difficult task of distinguishing between pictures of real objects and of artificial objects formed by combining parts from different real objects, even though they only seem to have partial access to semantic information. As Ellis and Humpreys (1999, p. 558) pointed out, these findings are problematical for other models: "Models such as those of Farah and McClelland (1991), which do not separate different forms of stored knowledge, find it more difficult to account for such a pattern of dissociation in which one ability stays intact."

The model does not provide a convincing explanation of those patients who have poorer naming and access to semantic information with non-living than with living things (e.g., Warrington & McCarthy, 1994). However, Ellis and Humphreys (1999, p. 554) argued that, "the effects can be accounted for if the lesion is not global but more selective, affecting the stored units and connections for the representations of non-living relative to living things."

General theory of high-level vision

Kosslyn et al. (1990) put forward an ambitious theory of high-level vision (visual processing involving the use of previously stored information). Evidence about brain functioning was used in its formation, and a computer simulation model was constructed to consider what components are necessary for high-level visual processing. This computer simulation model was also used to consider the results of different kinds of damage to the visual system.

The outline of the theory is shown in Figure 4.11. There are various sub-systems within the overall visual perceptual system, and each of these sub-systems consists of a parallel distributed network. In terms of the flow of information, the starting point is information resembling that in Marr's (1982) sketch (i.e., edge. depth, and orientation information) being delivered to the visual buffer. There is more information available in the visual buffer than can be passed on to the later stages of visual processing, and so there needs to be an attention window to handle this problem.

One of the central assumptions of the theory is that the encoding of object information (i.e., "what" information) and of spatial information (i.e., "where" information) occurs in separate subsystems. There is much support for this assumption (see Chapter 3). It is assumed that the spatial information supplied to the spatial properties sub-system from the visual buffer is retinotopic (location is specified relative to the retina). One of the main features of this sub-system is to transform this retinotopic representation, in which location is represented relative to objects in space. The object properties sub-system identifies the nonaccidental properties of the input on the basis of edge, texture, colour, and intensity information in ways similar to those proposed by Biederman (1987). Kosslyn et al. (1990) left it open whether this sub-system produces viewpoint-centred or object-centred object representations.

The associative memory sub-system is responsible for integrating spatial and object information supplied by the spatial properties and object properties sub-systems. This information is compared against appropriate stored information in order to produce object recognition. This is an ongoing process: as spatial and object information accumulates in associative memory, a hypothesis of the object's identity is generated. Finally, top-down search tests the hypothesis. It can be used to look up in associative memory the properties the hypothesised object should have, or it can produce a shift in attention if this is needed for object recognition.

Computer simulation

The implications of damage to parts of the visual processing system were assessed by Kosslyn et al. (1990) using computer simulation. Two-dimensional stimulus arrays representing either a face or a fox were placed in the visual buffer, and limited arrays consisting of one-ninth of the original array were passed on via the attention window to the other sub-systems. Four different tasks were then given to the computer simulation program:

3. Are they the same? (for two pictures presented in succession)

4. What is here? (for two pictures presented together)

The most striking finding of the computer simulation was that many perceptual problems can be caused by several different kinds of lesion or damage. One example is visual agnosia (involving deficient ability to recognise visual objects in spite of intact naming and attentional abilities), which was defined by poor performance on the first task listed earlier combined with intact performance on the third task. There were 34 different types of damage that produced this particular deficit. In similar fashion, there is prosopagnosia (difficulties in face recognition, see next section), which was defined by being able to identify a face as a face (first task) but being unable to identify correctly which face it was (second task). This pattern of performance was produced by 16 different types of damage.

Why can some disorders of visual perception be produced in numerous different ways? The main reason is because of the interconnected nature of the visual processing system shown in Figure 4.11. For example, damage within the object properties sub-system means there is an impoverished output from that sub-system to the associative memory sub-system. As a result, the associative memory sub-system cannot function effectively, even though it may be intact.

Kosslyn et al. (1990) also discussed a condition known as simultanagnosia, in which only one object at a time can be perceived (see also Chapter 5). The computer simulation revealed that this condition arose only through partial damage to that part of the spatial properties sub-system responsible for producing a spatiotopic representation. It could thus be predicted that simultanagnosia should be rarer than most other forms of perceptual deficit, and that is indeed the case.


The theory proposed by Kosslyn et al. (1990) has three major strengths. First, it was the first theory to propose computational processing sub-systems underlying high-level vision that are in line with available knowledge of brain systems. Second, it provides a useful framework for cognitive neuropsychologists in their efforts to make theoretical sense of the data from brain-damaged patients. Third, the theory is one of the few in which attentional and perceptual phenomena are integrated.

On the negative side, the theory is at too great a level of generality. As a result, there is little clarification of the detailed processes operating within each sub-system. This lack of specificity is perhaps especially noticeable so far as associative memory and top-down search are concerned. In both cases, it is much clearer what is accomplished by the particular sub-system than how it is accomplished.

Business Correspondence

Business Correspondence

24 chapters on preparing to write the letter and finding the proper viewpoint how to open the letter, present the proposition convincingly, make an effective close how to acquire a forceful style and inject originality how to adapt selling appeal to different prospects and get orders by letter proved principles and practical schemes illustrated by extracts from 217 actual letter.

Get My Free Ebook

Post a comment