Perceptual organization in the tilt illusion

Area of Science:

• Computational neuroscience
• Visual perception
• Machine learning

Background:

• The tilt illusion demonstrates contextual influences on visual orientation perception.
• Existing models often lack a basis in natural scene statistics.
• The Gaussian Scale Mixture (GSM) model offers a robust framework for natural scene statistics.

Purpose of the Study:

• To analyze the tilt illusion using a neural population model grounded in natural scene statistics.
• To investigate the role of divisive normalization in contextual effects on visual perception.
• To explore the implications of the GSM model for biological vision.

Main Methods:

• Developed a neural population model for visual orientation perception.
• Incorporated the Gaussian Scale Mixture model to represent natural scene statistics.
• Simulated divisive normalization of neural unit activities by surround stimuli.

Main Results:

• The model successfully replicates attraction and repulsion phenomena observed in the tilt illusion.
• Neural tuning curves were shown to change due to divisive normalization.
• Population decoding accurately predicted perceptual outcomes.

Conclusions:

• Divisive normalization, informed by natural scene statistics, can explain contextual effects like the tilt illusion.
• The model provides a rigorous framework for understanding visual contextual phenomena.
• This approach can be generalized to other visual attributes and perceptual effects.