Predicting Eye Fixations on Complex Visual Stimuli Using Local Symmetry.

Area of Science:

• Visual perception
• Computational neuroscience
• Image analysis

Background:

• Traditional bottom-up models for predicting human eye fixations primarily rely on contrast features.
• The Itti, Koch, and Niebur saliency model, a prominent contrast-saliency model, demonstrates success but lacks precision for mirror-symmetrical forms.
• Human observers tend to fixate on the center of symmetrical forms, contrasting with border-focused predictions of contrast models.

Purpose of the Study:

• To propose and evaluate a novel saliency model that incorporates local mirror symmetry for predicting human eye fixations.
• To compare the predictive accuracy of the proposed symmetry-based model against a traditional contrast-based model.
• To investigate the role of symmetry in guiding early visual attention.

Main Methods:

• Development of a saliency model utilizing local mirror symmetry detection.
• Conducting an eye-tracking experiment with human participants viewing complex photographic images.
• Comparing fixation predictions from the symmetry model and a contrast-based model (Itti, Koch, and Niebur) against empirical eye-tracking data.

Main Results:

• The proposed symmetry model significantly improves the prediction of human eye fixations compared to the contrast model.
• The enhanced predictive accuracy of the symmetry model is observed across a diverse range of images, not limited to those with explicit symmetrical content.
• Early human eye fixations are strongly correlated with highly symmetrical areas within images.

Conclusions:

• Local mirror symmetry is a crucial and powerful predictor of human eye fixations.
• Incorporating symmetry into saliency models offers a more precise method for predicting visual attention.
• Symmetry can effectively predict the order in which humans attend to different parts of an image.