How does image noise affect actual and predicted human gaze allocation in assessing image quality?

Area of Science:

• Visual Perception
• Computational Neuroscience
• Image Processing

Background:

• Understanding human gaze allocation in natural vision is crucial for scene perception.
• Previous studies focused on high-quality images, leaving generalization to degraded inputs unclear.
• Investigating gaze behavior under image degradation is essential for realistic models.

Purpose of the Study:

• To examine how image noise affects human gaze allocation during scene perception.
• To compare the performance of visual attention models in predicting gaze on degraded scenes.
• To identify factors influencing gaze distribution under noisy visual conditions.

Main Methods:

• Eye-tracking and computational analysis of human participants viewing distorted man-made and natural scenes.
• Systematic image degradation using Gaussian low-pass, circular averaging filters, and additive white noise.
• Comparison of four leading visual attention models against human gaze data.

Main Results:

• Distorted images elicited fewer fixations, longer fixation durations, shorter saccades, and stronger central fixation bias compared to clear images.
• The impact of noise on gaze was primarily driven by noise intensity, not the type of distortion.
• Natural scenes showed a more pronounced effect of noise on gaze than man-made scenes.
• Visual attention models performed worse than human inter-observer variance and lacked sensitivity to noise variations.
• Central fixation bias increased with noise intensity and was a strong predictor of human gaze.

Conclusions:

• External noise intensity critically determines scene-viewing gaze behavior in humans.
• Current visual attention models need refinement to account for noise effects observed in human vision.
• Future models should incorporate noise intensity as a key factor for realistic gaze prediction.