Contour integration and segmentation with self-organized lateral connections

Area of Science:

• Neuroscience
• Computational Vision
• Computational Neuroscience

Background:

• Contour integration in low-level vision relies on lateral interactions between neurons with similar orientation tuning.
• The developmental origin of these neural interactions has remained an open question in visual neuroscience.

Purpose of the Study:

• To propose a computational model explaining how lateral interactions for contour integration arise through self-organization.
• To demonstrate how synchronized neural firing represents contour perception and accounts for human performance.
• To investigate how input distribution during development influences contour integration performance across the visual field.

Main Methods:

• Development of a computational model simulating neural interactions in the visual cortex.
• Utilizing principles of input-driven self-organization to learn neural connections.
• Analysis of model performance in contour integration tasks and comparison with human psychophysics.

Main Results:

• The model demonstrates that lateral interactions for contour integration can be learned via self-organization, mirroring developmental processes in the visual cortex.
• Synchronized neural firing within the model effectively represents contour perception, achieving performance comparable to human capabilities.
• Model predictions show that contour integration performance varies across the visual field based on developmental input distributions.

Conclusions:

• The study grounds contour integration, a key perceptual phenomenon, in specific neural mechanisms.
• The model provides a framework for measuring structural and functional properties related to contour integration.
• The findings offer testable predictions to guide future experimental research in visual neuroscience.