Normalization by orientation-tuned surround in human V1-V3.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Perception

Background:

• The normalized energy model is a prominent theory explaining neuronal responses in the primary visual cortex.
• This model involves linear filtering, contrast energy extraction, and divisive normalization by a local population of complex cells.

Purpose of the Study:

• To investigate the limitations of the classical normalized energy model in explaining visual responses.
• To propose and validate a modified divisive normalization model that accounts for differential responses to various contour types.

Main Methods:

• Utilized human functional magnetic resonance imaging (fMRI) to measure neural responses.
• Implemented and tested a novel "tuned normalization" model where orientation-tuned cells preferentially inhibit each other.
• Validated the model's predictions against fMRI data for gratings and snakes, and generalized to other textures in V1 and extrastriate cortex (V2, V3).

Main Results:

• The classical normalized energy model failed to predict the observed fMRI responses, underestimating responses to curved contours (snakes) compared to straight contours (gratings).
• The proposed "tuned normalization" model successfully accounted for the differential fMRI responses between gratings and snakes.
• The model demonstrated generalization to other band-pass textures in both primary (V1) and extrastriate (V2, V3) visual cortices.

Conclusions:

• The classical divisive normalization model does not fully capture neural responses in the primary visual cortex.
• A modified "tuned normalization" model, incorporating orientation tuning specificity, provides a better account of neural processing.
• Image features like contour heterogeneity significantly impact neural responses, even in early visual areas.