Abstract—Quantifying visual attention in tourism promotional videos is crucial for enhancing digital communication efficiency. To enhance digital communication efficiency in the highly competitive global tourism marketplace, this research explores how AI-assisted visual analysis can optimize audience engagement. This research proposes a computational analysis framework utilizing a Convolutional Neural Network (CNN) integrated with the DeepGaze II architecture to predict saliency patterns in Thai tourism advertising campaigns. Visual features are extracted at the frame level using technical metrics, including the Weber contrast ratio (ΔL/L), optical flow magnitude, and Hue, Saturation and Value (HSV) color histograms. The system classifies viewer engagement into three attention tiers—High Attention (HAT), Medium Attention (MAT), and Low Attention (LAT)—based on simulated gaze density. Empirical results indicate that compositions with a high Weber contrast (ΔL/L > 1.5) and complex cultural symbols significantly correlate with attention scores exceeding 0.6. Furthermore, rhythmic pacing analysis reveals that alternating saliency levels contribute to a 19–23% increase in attention retention. Additionally, age-stratified gaze simulations demonstrate distinct responses to motion energy across different demographic groups. This study establishes an empirical link between visual formalism and computer science, providing a predictive tool for optimizing high-impact digital media design.

 

Keywords—saliency prediction, visual attention modeling, weber contrast, optical flow, image analysis

Cite: Ping Huang, Ratanachote Thienmongkol, and Ruethai Nimnoi, "AI-driven Saliency Modeling for Quantifying Visual Engagement in Tourism Media," Journal of Image and Graphics, Vol. 14, No. 4, pp. 720-732, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).