Natural scene statistics at the centre of gaze

Area of Science:

• Visual Neuroscience
• Computational Vision
• Human Psychophysics

Background:

• Early visual processing likely leverages natural visual stimuli structure.
• Active visual sampling (eye movements) may create unique stimulus statistics compared to passive viewing.
• Foveal and parafoveal vision receive high-resolution sampling and disproportionate cortical resources.

Purpose of the Study:

• To investigate how active eye movements influence the statistical properties of natural scenes sampled by the human visual system.
• To determine if active selection affects stimulus characteristics beyond the fovea.
• To explore the potential information-theoretic benefits of these selection effects.

Main Methods:

• Recorded human eye positions during natural scene viewing.
• Analyzed the spatial contrast and pixel correlation statistics of sampled image regions.
• Compared statistics of actively sampled regions with randomly selected image ensembles.

Main Results:

• Subjects preferentially fixated on image regions with higher spatial contrast.
• Within these high-contrast regions, pixel intensity correlations were reduced up to 4 degrees of eccentricity.
• These observed effects can be replicated by selecting only the highest-contrast image regions.

Conclusions:

• Active visual sampling significantly alters the statistics of encountered stimuli, favoring high-contrast areas.
• Reduced pixel correlations in these selected regions may enhance information transmission to the visual system.
• The visual system's active selection strategy appears optimized for efficient information extraction from natural scenes.