Agnosia may be defined as an impairment in the higher visual processes necessary for object recognition, with relative preservation of elementary visual functions. Studies in agnosia help provide both ideas and constraints to theories of the process of visual recognition. However, there is much debate on theories of visual recognition. There is also much controversy on how agnosia should be classified. These classifications often reflect different understandings of the process of visual recognition. Two recent classifications will be discussed in terms of their own hypotheses of visual recognition.

Farah (1995) distinguishes between apperceptive and associative agnosia. Apperceptive agnosia is thought to reflect any failure in object recognition in which perceptual impairments are clearly at fault, despite relatively preserved elementary visual functions. Farah proposes four subdivisions within apperceptive agnosia. The first category is apperceptive agnosia (narrow sense). These patients show adequate elementary visual functions, yet are impaired on the simplest form of shape discrimination. They cannot recognise, match, copy, or discriminate both simple stimuli and complex objects. The next category is labelled dorsal simultanagnosia. These patients have full (or near full) visual fields but can usually see only one object at a time. They have difficulties in localising an object in space, however, as long as they can see an object, they can recognise it. Ventral simultanagnosia is similar to dorsal simultanagnosia as patients are able to recognise only single objects. However, even though they cannot recognise multiple objects they can see them. The last category described by Farah is a perceptual categorisation deficit. These patients have great difficulties in matching three-dimensional objects across shifts in perspective and therefore have problems in naming objects when viewed from unusual perspectives.

In comparison, associative agnosia is thought to be due to problems in accessing semantic knowledge about an object even though the perceptual representation of the object is intact. Farah describes several subtypes of associative agnosias. Firstly, associative visual object agnosia (narrow sense) describes patients who have difficulty in recognising visually presented objects as seen though inabilities to name visually presented objects both verbally and non-verbally, and problems grouping objects according to their semantic category. These patients do, however, show normal recognition of objects though modalities other than vision and also retain the ability to copy objects or drawing and match similar pairs of stimuli. The next category of associative agnosia is prosopagnosia which reflects an inability to recognise faces. Another category, pure alexia, includes patients that cannot read normally despite apparently normal visual capabilities and preservation of writing and other language abilities. Farah also distinguishes other categories of agnosia such as having a selective deficit for living or nonliving things, or other category specific impairments.

Farah interprets both apperceptive and associative agnosia in terms of several theories of normal visual recognition. However, she concludes by advancing her own hypotheses of visual recognition. She postulates that higher vision begins with the grouping of local elements of the visual field into larger-scale contours, regions, and/or surfaces. Patients with apperceptive agnosia (narrow sense) are examples of a disruption in these grouping processes. They have perception of local contour, colour, and brightness yet have severe impairments of overall form. This stage of grouping separately registered local elements into higher order geometric representations is a prerequisite for virtually all higher visual processing and is a purely stimulus-driven process. The array of grouped stimulus elements interacts with two higher-level systems: the spatial attention system and the object recognition system. A disruption in the spatial system results in dorsal simultanagnosia, in which the contents of the grouped array that can be attended to in a given amount of time are abnormally limited. In contrast, a disruption in the object recognition system results in ventral simultanagnosia, in which the contents of the grouped array that can be recognised in a given amount of time are abnormally limited. The spatial attention system selects portions of the grouped array, and thus stimuli in this portion of the array are likely to be detected and recognised by the object recognition system.

The object recognition system redescribes portions of the grouped array in a more abstract format and captures the invariant 3-D geometric properties more fully than the grouped array. All objects are composed of parts which have certain spatial relations, and this system decomposes most objects into their constitute parts. These parts are then matched against object representations which results in the recognition of the object. This has led Farah to suggest that there are two underlying types of recognition ability being damaged in associative agnosia. Firstly, the ability to represent parts themselves, including complex parts for objects that undergo little decomposition. Secondly, the ability to rapidly encode multiple parts especially for objects that undergo much decomposition. Impairments in the ability to represent parts reflects the range of associative agnosias. That is, the more the representation of parts is impaired, the greater the recognition deficits. This can explain differences between prosopagnosia, and other object agnosias, as with severe deficits only the simplest objects will be recognised. In contrast, if the ability to represent objects is intact but the ability to rapidly encode multiple parts is impaired, then most objects will be recognised. Only objects which undergo decomposition into many parts will be affected. This type of deficit will result in alexia.

In contrast, Humphreys and Riddoch (1987) propose a different classification system which they believe reflects the complexity of the component processes involved in visual object recognition, and the manner in which such processes may selectively break down. They propose that there are six types of deficits that may be applied to patients with visual agnosia. These include impairments in shape processing, transformation processes, integration processes, access to form knowledge, access to semantics, loss of stereoscopic vision, and impaired semantic knowledge. The impaired shape processing classification refers to patients whose recognition problems are attributable to deficits in processing the primitive characteristics of shape such as edges at different orientations and spatial scales. Impaired transformation processes involve a classification of patients that can perform simple shape matching tasks, identify objects from a prototypical view, but are unable to identify objects from an unusual view. In contrast, patients with loss of access to form knowledge typically have difficulty in accessing or using stored information about visual form. Patients may show difficulties in drawing from memory yet will be good at copying.

Impaired access to semantics reflects the next category of visual agnosia. These patients have access to the object form system as well as intact semantic knowledge, however, they are impaired in accessing knowledge visually. However, as their semantic knowledge is intact, they can access this knowledge via another modality. Loss of stereoscopic vision constitutes another classification. Patients with this problem are able to derive well integrated 2-D representations of objects but fail to assign depth to these representations. This deficit impairs object recognition when recognition is dependent upon the derivation of accurate surface descriptions. The last classification involves impaired semantic knowledge. Patients show no deficits on pre-semantic visual object processing yet are unable to recognise these objects. Unlike patients with semantic access impairment, semantic agnostic patients have damage to the semantic system and thus cannot recognise objects via other modalities.

This type of classification is directly reflected by an information processing framework proposed by Humphreys and Riddoch. This model begins with a visually presented object. The first two processes of visual recognition are local geometric feature processing and global shape processing. They propose that if a lesion occurs at the global shape processing site, shape processing will be impaired. Information from these first two processes help to form a viewpoint-dependent object description. A lesion at this site results in impaired 2-D shape integration or the coding of integrated shape descriptions. Information from all three processes so far form an abstract episodic object description. Damage to the abstract episodic object description will impair transformation processes. Information from both the abstract episodic object description and the viewpoint dependent object description then access object form knowledge. Damage to the object form knowledge site will result in impairments in accessing knowledge of visual form. The next stage is semantic knowledge of the object, and damage will impair semantic knowledge of all objects in all modalities. Object form knowledge and semantic knowledge send information to one another and lesions in the flow of knowledge to one another will affect access to semantic knowledge, however, semantic knowledge is spared thus it may be accessed via other modalities. It is evident that the classification system of Humphreys and Riddoch is derived directly from their model of visual recognition and the deficits that can occur at each stage of the model.

Both classifications described are based on vastly different models. Farah’s model of recognition is much better in explaining dissociations between different types of associative agnosias such as prosopagnosia, pure alexia, and other selective object agnosias. In contrast, Humphreys and Riddoch’s information processing model of visual recognition is much more open to further expansion in a way in which Farah’s apperceptive-associative distinction is not. For example, it is becoming increasingly clear that perception is not entirely normal in the vast majority of associative agnosia, thus, not all types of agnosia can be easily categorised into Farah’s broad classifications of apperceptive and associative agnosia.

In conclusion, there is currently much debate over the processes underlying visual recognition. Both classifications discussed are based on different models, and it is evident that each of the disorders in both classification systems illuminate disruptions of different stages of normal visual recognition. Further research of visual agnosia will increase knowledge of normal recognition. Subsequently, a greater understanding of normal visual recognition will help researchers devise new and better ways of classifying agnostic patients.