A dynamical scan-path model for task-dependence during scene viewing.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Computer Vision

Background:

• Human observers use sequences of fixations for scene perception.
• Visual attention models predict gaze positions using saliency maps.
• Scan path models are evolving to include constraints and task influences.

Purpose of the Study:

• To propose a novel scan-path model incorporating fixation durations.
• To account for task instructions and interindividual differences in viewing behavior.
• To dynamically model spatial-temporal gaze patterns.

Main Methods:

• An eye-movement experiment with four task conditions.
• A fully Bayesian dynamical modeling framework.
• Joint spatial-temporal likelihood with sequential estimation.

Main Results:

• Model parameters were estimated for individual observers and tasks.
• Attentional span adjusted to specific task requirements.
• The dynamical model reproduced task-specific scan-path statistics.

Conclusions:

• The proposed model effectively integrates task demands into scan path prediction.
• It offers a more nuanced understanding of visual attention and eye movements.
• This framework advances computational models of real-world scene perception.