Visual causes versus correlates of attentional selection in dynamic scenes

Area of Science:

• Cognitive Neuroscience
• Visual Perception
• Computational Vision

Background:

• Understanding attentional selection is crucial for explaining visual perception.
• Distinguishing between causal factors and mere correlates of attention is a key challenge.

Purpose of the Study:

• To investigate the causal visual features driving attentional selection.
• To compare the predictive power of various bottom-up visual models in naturalistic viewing conditions.

Main Methods:

• Eye-tracking experiments were conducted using dynamic video clips.
• Seven bottom-up visual models (intensity variance, orientation contrast, intensity contrast, color contrast, flicker contrast, motion contrast, integrated saliency) were evaluated.
• Saccade target selection predictions were quantified for each model.

Main Results:

• All tested models predicted saccade target selection above chance.
• Dynamic visual models (flicker, motion contrast) were highly predictive, especially for early, bottom-up driven saccades.
• Static models showed weaker predictive power, with intensity variance and orientation contrast being particularly poor predictors.

Conclusions:

• Dynamic visual cues play a dominant causal role in attracting attention.
• Some static visual cues have a weaker causal influence, while others may not be causal at all.
• These findings highlight the importance of dynamic information in guiding visual attention.