Time-dependent modulation of saliency-driven visual attention in free viewing.

PURPOSE: This study investigated similarities in gaze behaviour across observers and examined whether salient visual features guided visual attention during free viewing, to clarify bottom-up and top-down influences on visual attention in complex, natural scenes.

METHODS

Forty-five healthy adults (Median age (IQR):38(24)y) viewed ten static real-world scenes (with/without social elements) for 15 s each while their gaze was simultaneously recorded using a remote eye-tracking system. Fixation heatmaps were generated, and gaze parameters were computed. Using a bootstrapping approach, random subsets of gaze data were compared to assess gaze similarity. Heatmap similarities were analysed during the first 7.5 s (tearly), last 7.5 s (tlate), and total 15 s viewing period using Jensen-Shannon Divergence (JSD), where JSD= 0 indicated identical maps, and JSD= 1 indicated no similarity. A saliency map for each scene was obtained using Itti & Koch's computational model, and the proportion of fixations overlapping salient regions was calculated to quantify saliency's influence during free viewing.

RESULTS

Across the total viewing period, 52-84% of fixations overlapped with model-predicted salient regions, with higher proportions observed in less crowded scenes and lower proportions in scenes containing socially meaningful elements. Temporal comparison showed a higher proportion of saliency-directed fixations during the tearly phase [median (IQR): 72% (23%)], accompanied by greater inter-group gaze similarity [median (IQR) JSD: 0.15 (0.02)], indicating more shared, saliency-driven viewing during the first 7.5 s.

CONCLUSIONS

The data suggest that free viewing is initially consistent and saliency-driven but becomes progressively individualised over time as top-down influences increase, with this temporal shift modulated by scene characteristics. Free-viewing can hold clinical value in assessing the dynamic interaction between bottom-up saliency and top-down cognitive context in individuals with visual or cognitive impairments.