Abstract
Since Neisser (1967) and Treisman & Gelade (1980) many models of selective visual attention have included a preattentive stage of processing. Some researchers have suggested that the term “preattentive” has outlived its usefulness. While there are uses of “preattentive” that should be avoided, we argue against abandoning the concept altogether. If there is selective visual attention and if there is processing of visual stimuli prior to selection, then there is preattentive processing. We hold that a goal of this processing is to abstract a preattentive representation from the stimulus that can be used to guide attention. In that representation, features are coarsely coded in parallel. It is a mistake to think of this representation as self-sufficient “preattentive vision” whose output can directly control behavior. While some forms of perceptual grouping and segregation may be accomplished preattentively, it does not seem to be possible to go directly to a motor response without engaging attention-demanding processes. In particular, while there may be some preattentive conjoining of features, the accurate explicit binding that appears to be required for object recognition entails selective attention to specific objects. Building on the Reverse Hierarchy Theory of Hochstein & Ahissar and on re-entrant ideas of DiLollo, we suggest that features are initially encoded through a feed-forward pathway from V1 to higher visual areas in both ventral and dorsal pathways. Preattentive guidance of focused attention can then be modeled as re-entry from anatomically high-level loci onto earlier loci. Parietal areas appear to guide spatial selection and prestriate areas may guide feature-based selection. In this framework, no part of the brain is exclusively “preattentive”. The same anatomical areas may participate in the feed-forward production of a preattentive representation and then in the more detailed, selective processing modulated by attention through subsequent reentry.