Glimpse: A Gaze-Based Measure of Temporal Salience

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Cognitive Science

Background:

• Visual salience research has predominantly focused on spatial aspects, neglecting the critical temporal dimension of attention.
• Understanding how attention evolves over time is crucial for analyzing dynamic visual content.

Purpose of the Study:

• To introduce Glimpse, a novel computational measure for quantifying temporal visual salience.
• To provide a gaze-based method for analyzing the temporal dynamics of visual attention.

Main Methods:

• Developed Glimpse, a measure based on observer-spatio-temporal consistency of raw gaze data.
• Explored scoring algorithms for estimating temporal salience from existing spatial salience models.

Main Results:

• Glimpse offers a simple, training-free quantification of temporal visual attention.
• Gaze-based temporal salience measurement significantly outperforms methods relying on predicted spatial salience maps.

Conclusions:

• Glimpse effectively captures the temporal evolution of visual attention in dynamic scenes.
• The proposed measure has potential applications in video segmentation and summarization, with publicly available software and data.