Abstract
We present a theory of visual attention motivated by recent, seemingly contradictory psychophysical findings. We previously observed that natural scenes and objects could be categorized while attention was engaged by a concurrent task, suggesting that not only simple features can be processed “preattentively”. Yet the same natural scenes or objects were not necessarily categorized in parallel during visual search experiments, implying that the two paradigms (dual-task and visual search) might in fact reveal distinct attentional resources. We proposed that the ability to discriminate stimuli preattentively (in dual-task) is dependent upon the existence of selective neuronal populations at any level of the visual cortical hierarchy, while parallel visual search (which is expected to occur for such stimuli) can sometimes be prevented by competition within the large receptive fields (RFs) of neurons in high-level visual areas. Indeed, we found that in such cases, increasing inter-stimulus distance could facilitate visual search. We organize these and other findings into a unified framework by proposing that (i) the preattentive visual system can be equated with feed-forward cortical selectivities, and (ii) under the common label “attention” are grouped at least two collections of feedback processes. The first one corresponds to “feature binding” mechanisms, allowing to process objects for which no selective neuronal population exists. The other corresponds to “biased competition” mechanisms, with the role of combating RF clutter. This framework thus reconciles two “classical” theories of attention. By relying on common properties of neuronal systems, such as cortical selectivities, hierarchical organization, feed-forward and feedback circuitry, and RF sizes, this theory explains the apparent discrepancy between the remarkable efficiency of our visual systems on the one hand, and its dramatic failures observed in situations such as change blindness, on the other hand.