A recurrent neural model for proto-object based contour integration and figure-ground segregation

Area of Science:

• Computational Neuroscience
• Visual Perception
• Cognitive Neuroscience

Background:

• Visual object recognition involves feedforward and feedback information streams.
• The precise role and targets of feedback signals in lower visual areas remain unclear.
• Understanding these mechanisms is crucial for explaining contour integration and figure-ground segregation.

Purpose of the Study:

• To develop a recurrent neural model explaining feedback mechanisms in visual processing.
• To investigate the role of grouping neurons in representing tentative objects ('proto-objects').
• To elucidate how Gestalt principles influence visual scene interpretation via neural inhibition.

Main Methods:

• Developed a recurrent neural network model incorporating local feature neurons and grouping neurons.
• Modeled feedback projections from grouping neurons to local feature neurons.
• Incorporated inhibitory mechanisms at local feature and object representation levels.

Main Results:

• The model successfully explains existing neurophysiological data on visual processing.
• It demonstrates how grouping neurons integrate local features to form proto-objects.
• The model predicts the impact of feedback and attention on neural activity.

Conclusions:

• Feedback signals play a crucial role in object-based visual processing.
• The model provides a framework for understanding object-based attention.
• This work clarifies the neural basis of contour integration and figure-ground segregation.