An inference upon the neural network finding binocular correspondence

Area of Science:

• Computational neuroscience
• Computer vision
• Visual perception

Background:

• Previous work introduced a neural network model for extracting binocular parallax.
• Binocular depth neurons are sensitive to specific parallax but can misfire with incorrect pairs.

Purpose of the Study:

• To propose a neural network model that identifies correct binocular correspondence.
• To explain how the visual system resolves ambiguity in binocular matching.

Main Methods:

• A cascaded neural network model was designed to process outputs from binocular depth neurons.
• The model's performance was evaluated through digital computer simulations.

Main Results:

• The proposed model successfully identified correct binocular correspondences.
• Computer simulations demonstrated the model's ability to explain hysteresis in binocular depth perception.

Conclusions:

• The developed neural network effectively solves the binocular correspondence problem.
• The model provides a computational explanation for perceptual phenomena like hysteresis.