Abstract
Visual selective attention limits processing resources to some items at the expense of others. Typical models of selective attention have had a linear sequence of stages in which early vision extracts features that can guide search and support texture segmentation. Subsequently, an attentional bottleneck restricts access to limited-capacity object recognition processes. Three findings demand a change in the model. First, double dissociations between efficient search and effortless texture segmentation undermine the idea of a set of preattentive features common to both tasks. Second, double dissociations between efficient search and item identification under dual task load (VanRullen et al.) implausibly imply that some identifications both require and don't require attention. Finally, some features (e.g. T-junctions, intersections) are available to early vision and to later object recognition process. However, these features do not guide attention in search. In a linear model, how can features be available before and after the bottleneck while being unavailable at the bottleneck? Our new, two-pathway architecture addresses these and other issues. The older, two-stage model is captured in a SELECTIVE PATHWAY running from early vision through a bottleneck to later object recognition processes. Access to the bottleneck is controlled by a GUIDING REPRESENTATION abstracted from the input. We conceive of this as a control system sitting to one side of the selective pathway. Search efficiency is governed by the properties of the selective pathway and guiding representation. A second, NON-SELECTIVE PATHWAY mediates visual processes not subject to a selective bottleneck (e.g. aspects of texture processing and some visual discriminations). The two pathways approach can move discussion beyond debates about “preattentive”, “serial”, and “parallel” processing. Relationships with other two-pathway models (e.g. Ungerleider & Mishkin, 1982 and Goodale, 1996) will be discussed.
Supported by NIH-NEI and AFOSR