Predicting cortical dark/bright asymmetries from natural image statistics and early visual transforms

Area of Science:

• Neuroscience
• Computational Vision
• Image Analysis

Background:

• Sensory systems leverage environmental predictability for efficient encoding.
• Previous research identified non-uniform distributions of visual features in natural images.
• Statistical distributions of visual features differ between bright and dark image components.

Purpose of the Study:

• To investigate how statistical differences in bright and dark image components influence visual cortical input.
• To explore the role of retinal ON and OFF pathways in processing these differences.
• To provide a potential explanation for observed asymmetries in visual feature processing.

Main Methods:

• Analysis of statistical distributions of visual features (contrast, scale, depth) in natural images.
• Modeling of early visual pathways, including magnocellular (M) and parvocellular (P) pathways with ON/OFF divisions.
• Transformation of natural images into statistical patterns of cortical input using pathway models.

Main Results:

• Significant differences in statistical distributions of visual features between bright and dark image components were confirmed.
• These differences translate into distinct patterns of cortical input from retinal ON and OFF pathways.
• Identified dark/bright asymmetries in cortical input align with recent findings on visual processing asymmetries.

Conclusions:

• Natural image statistics, specifically the dark/bright dichotomy, create regularities in visual cortical input.
• These input asymmetries offer a potential explanation for perceptual asymmetries in visual feature processing.
• The dark/bright distinction is a key factor in generating both cortical and perceptual asymmetries.