Modelling saliency attention to predict eye direction by topological structure and earth mover's distance

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Predicting eye direction is crucial for understanding user attention and behavior.
• Existing saliency attention models often rely on Difference of Gaussian (DoG) for feature extraction.

Purpose of the Study:

• To propose a novel saliency attention model for enhanced eye direction prediction.
• To leverage topological structure and Earth Mover's Distance (EMD) for improved saliency feature extraction.

Main Methods:

• Extract visual saliency features (color contrast, intensity contrast, orientation, texture).
• Preserve topological structure by eliminating disconnected regions in feature maps.
• Utilize across-scale EMD for center-surround difference calculation, replacing DoG.
• Employ across-scale fusion of feature maps and a competition function for saliency map generation.

Main Results:

• The proposed model effectively extracts and fuses visual saliency features.
• The model demonstrates superior performance in eye direction prediction compared to existing methods.
• Experimental results validate the efficacy of the EMD-based approach.

Conclusions:

• The developed saliency attention model achieves state-of-the-art performance in eye direction prediction.
• The integration of topological structure and EMD offers a promising direction for attention modeling.
• This approach has significant implications for human-computer interaction and user behavior analysis.