Guided Search 2.0 A revised model of visual search

Wolfe, Jeremy M. · 1994 · OpenAlex-citations

DOI: 10.3758/bf03200774

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper introduces Guided Search 2.0 (GS2), a revised computational model of visual search designed to explain how humans efficiently locate specific items within cluttered visual fields. The research addresses the fundamental problem of how the visual system manages limited processing capacity. The author argues that while early visual processing is massively parallel and preattentive, extracting basic features like color and orientation, subsequent identification processes are limited in capacity and must be deployed serially. The core hypothesis of GS2 is that this deployment of attention is not random but is guided by the output of the earlier parallel feature processes, allowing the system to prioritize locations most likely to contain the target. The methodology involves a detailed computer simulation of the GS2 model, which is structured into two main stages. The first stage processes input through independent "feature maps" for basic attributes such as color and orientation. These maps are filtered through broadly tuned, categorical channels (e.g., "steep," "shallow," "red," "green"). The second stage generates an "activation map" by combining two types of activation: bottom-up (stimulus-driven) and top-down (user-driven). Bottom-up activation is calculated based on local differences between an item and its neighbors, thresholded by preattentive just noticeable differences (Pjnds), and weighted by channel response strength. Top-down activation is determined by selecting the specific feature channel that best differentiates the target from distractors, based on rules prioritizing unique categories. The simulation was tested against standard visual search paradigms, including reaction time and percent correct measures across varying set sizes. The results demonstrate that the GS2 simulation successfully reproduces human performance data for a wide variety of search tasks. Specifically, the model accounts for "parallel" searches, where reaction times are independent of set size for targets defined by a single unique feature (e.g., a red item among green distractors). It also explains "serial" searches, where reaction times increase linearly with set size for conjunction targets (e.g., a "T" among "L"s), including the characteristic difference in slope between target-present and target-absent trials. The simulation shows that the combination of bottom-up and top-down signals allows for efficient guidance even when the target is not uniquely defined by a single feature, such as in conjunction searches where the target shares features with different distractors. The significance of this work lies in providing a unified, explicit framework that reconciles previous theories, such as Treisman’s Feature Integration Theory and Neisser’s preattentive/attentive distinction. By making the mechanisms of attentional guidance explicit, GS2 offers a more robust explanation for visual search efficiency than earlier models. The paper concludes by acknowledging current shortcomings, such as the arbitrary nature of some channel definitions and the lack of a dynamic feedback mechanism for channel selection, suggesting areas for future refinement. Overall, GS2 establishes that visual search is a guided process where parallel feature analysis directs limited-capacity attention, optimizing the speed and accuracy of object identification.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified_with_issues.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• visual search
• inattentional change blindness