Abstract
Guided Search (GS) is a model of visual search implemented as a computer simulation. In GS2 (1994), items were ordered in terms of the amount of preattentive guidance and search proceeded from most likely to least likely until the target was found or until an adaptive “activation threshold” was reached and search was terminated. However, since then, our data have suggested that the search mechanism does not keep track of rejected distractors. This and other new results require a new GS implementation. Following in the tradition of Treisman's Feature Integration Theory, GS4 has two main stages. The first, preattentive stage generates a guidance map to direct the deployment of attention in the second stage. The preattentive stage processes a limited number of basic features in parallel across the visual scene. Color, size, and orientation are implemented in the simulation. There are two forms of preattentive guidance. Bottom-up guidance directs attention toward objects whose features differ from their neighbors. Top-down guidance directs attention toward objects that have target features. The guidance map is the noisy, weighted sum of top-down and bottom-up components. In GS4, target identification requires that preattentive features (e.g. red, vertical) must be bound by attention into a coherent object (e.g. a red, vertical bar) and that binding requires attention. In GS4's second, attentive stage, attention is deployed to the item with the greatest current preattentive activation, without regard to the previous history of deployments. GS4 successfully mimics human RT and error data for feature, conjunction, and “serial” search.It shows search asymmetries and eccentricity effects. It does not require memory for rejected distractors. Finally, it makes testable predictions about new search phenomena some of which we have been able to confirm.