A bottom-up model of spatial attention predicts human error patterns in rapid scene recognition

Area of Science:

• Cognitive Neuroscience
• Visual Perception
• Human Factors

Background:

• Humans excel at detecting targets in dynamic natural scenes, often with minimal focal attention.
• The specific factors limiting trial-by-trial detection performance remain unclear, distinguishing between bottom-up visual processing and top-down influences.

Purpose of the Study:

• To investigate the relative contributions of visual processing and top-down factors in rapid target detection.
• To determine if attentional limitations or target recognition primarily affects performance in dynamic visual sequences.

Main Methods:

• Eight observers performed an animal detection task with natural scenes presented at 20 Hz.
• Statistical stimulus properties (contrast, 'surprise') were analyzed for trial-by-trial performance prediction.
• A second experiment reordered sequences to manipulate 'surprise' and assess its causal impact on detection.

Main Results:

• Trial-by-trial performance was consistent across observers, indicating limitations in visual processing rather than individual variability.
• 'Surprise' in adjacent images and target contrast predicted detection success.
• Increased 'surprise' due to sequence reordering significantly impaired target detection.

Conclusions:

• Attentional limitations, specifically influenced by stimulus 'surprise', play a crucial role in rapid target detection.
• Findings challenge previous notions by highlighting attention's impact over mere target recognition in dynamic visual tasks.