Abstract
High-level perceptual aftereffects have assisted in resolving the long-standing question of whether the dimensions of faces are encoded as deviations from a statistical average (norm-based coding) or as absolute values (exemplar-based coding). However, this technique has not been used in the investigation of the processing strategies underlying non-face object perception. The present research investigated norm-based versus exemplar-based coding of horses (which, like faces, have a shared first-order configuration), houses (which can vary dramatically in their first-order configuration), and faces. Horses were distorted by changing their stockiness, houses by changing the window size, and faces by expansion or contraction. Norm-based (broadband-opponent) models predict that the magnitude of aftereffects will increase with distance of the adaptor from the proposed norm, and that adaptation to any stimulus that is not the norm will produce aftereffects that affect perception of the whole range of stimuli. In contrast, exemplar-based (multiple-narrowband) models predict the strongest aftereffects for test stimuli close to the adaptor, with aftereffects rolling off as the distance from adaptor to test stimulus is increased. Horses, like faces, produced aftereffects consistent with a norm-based) processing strategy. Results for houses were inconsistent with norm-based coding, and more consistent with an exemplar-based representation. The results are discussed in terms of the uniqueness of face processing, and the importance of a shared first-order configuration and observer experience as a determinant of the processing strategies used by the visual system.
Supported by Australian Research Council grants DP0558816, DP0450636, and DP0984558.