Guiding attention of faces through graph based visual saliency (GBVS).

Area of Science:

• Computer Vision
• Cognitive Science
• Human-Computer Interaction

Background:

• Human attention is naturally biased towards certain faces due to dominant perceptual features.
• Visual saliency, or the 'salient face', is determined by feature dissimilarity among surrounding faces.
• Modeling computer vision systems to emulate human visual processing is an active research area.

Purpose of the Study:

• To propose a novel graphical, bottom-up approach for identifying salient faces in images with multiple faces.
• To computationally model the human tendency to focus on specific faces in a crowd.
• To enhance intelligent vision systems by replicating human-like visual attention.

Main Methods:

• A graphical, bottom-up method was developed to calculate visual saliency of faces.
• Saliency was determined using facial intensity values, areas, and relative spatial distances.
• Experiments were conducted on grayscale images, with validation against ground truth, saliency scores, and standard parameters.

Main Results:

• The proposed method successfully identifies salient faces, with predictions aligning with human visual perception in some aspects.
• Validation across three levels confirmed the reliability and effectiveness of the saliency maps.
• The approach demonstrated moderately improved results compared to existing methods.

Conclusions:

• The developed graphical approach effectively identifies salient faces by considering visual features and spatial relationships.
• The findings suggest that this method can be valuable for future development of intelligent vision and robot vision systems.
• The study highlights the potential of computational models to mimic human cognitive processes in visual attention.