Local interactions in neural networks explain global effects in Gestalt processing and masking

Area of Science:

• Visual perception research
• Computational neuroscience
• Psychophysics

Background:

• Object segmentation and formation are key questions in vision research.
• The shine-through effect demonstrates how target visibility depends on the masking stimulus's spatiotemporal Gestalt.
• Previous findings suggested high-level Gestalt principles, like homogeneity, were necessary for explanations.

Purpose of the Study:

• To investigate the underlying mechanisms of the shine-through effect.
• To determine if low-level properties, rather than global Gestalt, can explain shine-through.
• To model the shine-through effect using a neural network.

Main Methods:

• Utilized the shine-through paradigm with a vernier target preceding a grating mask.
• Varied the spatial extent and homogeneity of the masking grating.
• Developed and analyzed a Wilson-Cowan type neural network model.

Main Results:

• Vernier visibility was dependent on the grating's size and homogeneity.
• Subtle deviations from homogeneity abolished the shine-through effect.
• The neural network model qualitatively and quantitatively replicated the observed effects.
• Model visibility correlated with transient neural population activation from local interactions.

Conclusions:

• The shine-through effect can be explained by local neural interactions within a network model.
• Low-level properties and local dynamics are sufficient to explain the shine-through effect.
• Explicit global Gestalt processing is not required for understanding this visual phenomenon.