Learning the Gestalt rule of collinearity from object motion

Area of Science:

• Neuroscience
• Computational Vision
• Developmental Neuroscience

Background:

• The Gestalt principle of collinearity is thought to arise from long-range connections in the primary visual cortex.
• These connections are traditionally considered innate or developed through early visual experience.

Purpose of the Study:

• To investigate the role of visual experience and object motion in developing long-range connections.
• To propose and test a neural network model for learning these connections.

Main Methods:

• Review of neurophysiological and psychophysical literature.
• Development of an unsupervised Hebbian learning neural network model.
• Utilizing spatiotemporal retinal filtering sensitive to visual input changes.

Main Results:

• The model successfully learned long-range connections from real camera sequences.
• Correlation of transient neural responses, not sustained ones, proved crucial for learning.
• Learning efficacy was highest with video sequences of moving objects.

Conclusions:

• Long-range connections underlying collinearity are learned from visual experience after birth, driven by coherent object motion.
• The model provides insights into the a priori knowledge necessary for self-organized structuring of the brain through experience.