Impairments in part-whole representations of objects in two cases of integrative visual agnosia.
Visual object recognition is undoubtedly a complex process. This is testified by the many attempts to build robust computer vision systems that are capable of recognising objects across a range of views, under different lighting conditions and in different contextual environments (see Brady, 1997; Lowe, 1987 for examples). It is also a commonplace assumption in the computer vision community that visual recognition comprises a number of distinct steps-these include: edge extraction, grouping of local image features, segmentation of objects from the background and from other objects that may be present, formation of a structural description of the object, and accessing stored structural and semantic information from the description assembled from the image (e.g. see Biederman, 1987; Marr, 1982). If the brain follows similar steps to achieve recognition, and if these steps are at least to some degree localised in different neural regions, we might expect visual recognition to break down in a variety of ways, according to the nature of the component processes involved. In contrast to computational arguments for there being several necessary substages in object recognition, the neurological and neuropsychological literature on recognition disorders has traditionally adopted a dichotomous approach. This approach originates in the pioneering work of Lissauer (1890). Lissauer distinguished between two forms of recognition disorder or visual agnosia: apperceptive and associative. Apperceptive agnosia was diagnosed as an impairment that disrupts the formation of a normal percept for the visual stimulus,