Searching for meaning: Local scene semantics guide attention during natural visual search in scenes

Area of Science:

• Cognitive Psychology
• Neuroscience
• Computer Vision

Background:

• Visual search models often use image salience to predict attention.
• Scene meaning is correlated with image salience, raising questions about its role.

Purpose of the Study:

• To investigate whether scene meaning or image salience primarily guides visual attention during object search.
• To differentiate the influence of semantic informativeness versus conspicuous features on attentional guidance.

Main Methods:

• Generated 'meaning maps' and 'salience maps' for real-world scenes.
• Analyzed fixation densities from two object search tasks.
• Compared the variance in fixation densities explained by meaning versus salience.

Main Results:

• Scene meaning significantly predicted fixation densities more than image salience.
• Meaning explained 58% and 63% of the variance in attention across studies.
• Initial saccades, regardless of speed, were directed to meaningful regions over salient ones.

Conclusions:

• Visual attention during search is primarily driven by scene meaning, not just image salience.
• The visual system prioritizes semantically informative regions for attention, even in a task-neutral manner.
• This finding challenges models relying solely on salience for attentional guidance in complex scenes.