Visual search

Searching a visual display or scene for a target defined by a task. Response time, accuracy, eye movements and stopping behavior provide different constraints. Candidate set size, similarity, visibility, prevalence and scene context all matter.

Changes in search behavior and misses when targets become more or less frequent. A controlled artificial baggage-search study found more misses at low target prevalence, together with altered absent-decision times. The phenomenon motivates modeling decision criteria and stopping in addition to feature processing; laboratory rates are not automatically professional field rates.^[ref]

A model family in which initial visual information and task goals prioritize candidate objects for selective processing. Selection is not necessarily a random scan, and serial selection can coexist with overlapping recognition processes.

A 2021 modeling development by Jeremy M. Wolfe, combining feature-based guidance, selection history, reward and scene information with selection, recognition and quitting processes. It models serial deployment with overlapping processing rather than assuming that only one complete operation occurs at a time. Model timings are hypotheses tested by simulation, not direct anatomical measurements.^[ref]

Assume $N$ candidate items are examined once each, each costs time $\tau$, nonsearch processing costs $b$, and a uniformly positioned target ends search when encountered. The target rank has mean $N^{-1}{\sum }_{k=1}^{N}k=(N+1)/2$, proving the present-trial expression. Absent trials inspect all $N$ items. The resulting 2:1 slope ratio is a benchmark under these assumptions, not a unique diagnostic of a serial mechanism.

Search in which the target is distinguished by a particular combination of features that individually occur among distractors. Guidance and grouping can make such search more efficient than exhaustive random item-by-item inspection.

Search in which an individual target feature distinguishes it from distractors. Efficient performance can produce a shallow response-time/set-size relation, but that relation alone does not prove absence of attention or a strictly unlimited parallel mechanism.