Abstract
Traditional models of visual search propose (a)that visual search is driven by similarity such that attentional priority is given to those objects most similar to the target and (b)that attentional efficiency is indexed by the slope of the search function. Here we show evidence that both of these basic tenants of visual search are fundamentally wrong. Our theory (Information Theory of Vision; Lleras, Cronin & Buetti, submitted) proposes that visual search is the product of two sequential and independent stages: (1)an attentional screening stage, where information in the display is processed with the goal of determining the locations most likely to contain the target; and (2)a subsequent scrutiny stage, where those locations are scrutinized. Crucially, screening is resolution limited. It cannot find the target location when items sufficiently similar to the target (candidates) are in the display. However, it can readily discount sufficiently dissimilar items (lures). We show that most of visual search, in fact, happens during (1), not (2), both in traditional laboratory search tasks and in search through real-world scenes. Our theory proposes that the duration of (1) respects Hick's law: it is proportional to the information in the display, when information is measured according to Shannon's uncertainty principle. In four experiments, we present empirical evidence that: (1) is directly proportional to information, producing RT functions that increase logarithmically as a function of the number of items in the display; and (2) attentional scrutiny (the slope) is actually unaffected by candidate-lure dissimilarity. We estimated the parameters of our model based on these four experiments and used them to accurately predict performance (Rsq>0.90) in new experiments containing novel combinations of lures. The data also provide evidence that lures are rejected in sequential, decreasing order of dissimilarity to the target, quickly reducing the uncertainty about locations worthy of precise scrutiny.
Meeting abstract presented at VSS 2014