Free-energy and illusions: the cornsweet effect.

Area of Science:

• Computational Neuroscience
• Visual Perception
• Bayesian Inference

Background:

• Visual illusions challenge traditional perception models.
• Bayesian perception offers a framework for understanding how the brain infers sensory input.
• The free-energy formulation provides a principled approach to Bayesian inference.

Purpose of the Study:

• To review visual illusions through the lens of Bayesian perception.
• To demonstrate that illusory percepts are Bayes-optimal explanations for ambiguous sensory data.
• To model the Cornsweet effect and Mach bands using a generative model.

Main Methods:

• Utilized the free-energy formulation of Bayesian perception.
• Developed a generative model of visual scenes incorporating prior beliefs about illuminance and reflectance.
• Simulated neuronal processing via neurobiologically plausible inversion of the generative model.
• Verified model predictions psychophysically using brief visual stimuli and a forced-choice paradigm.

Main Results:

• The Cornsweet effect emerges as a natural consequence of Bayes-optimal perception.
• Mach bands appear as secondary illusory percepts dependent on stimulus contrast.
• The generative model successfully simulated contrast-dependent illusion emergence.
• Psychophysical verification confirmed the model's qualitative and quantitative predictions.

Conclusions:

• Illusory percepts are not errors but Bayes-optimal interpretations of ambiguous sensory information.
• Bayesian perception, particularly the free-energy formulation, provides a robust framework for understanding visual illusions.
• The developed model offers a neurobiologically plausible account of illusion generation and perception.