Dynamic decorrelation as a unifying principle for explaining a broad range of brightness phenomena

Area of Science:

• Computational neuroscience
• Visual perception
• Image processing

Background:

• The human visual system's sensitivity to spatial context is crucial for interpreting luminance patterns.
• Numerous neural mechanisms have been proposed to explain brightness perception, driven by context sensitivity.
• Visual illusions present significant challenges to understanding brightness perception.

Purpose of the Study:

• To propose a novel computational model for estimating brightness in visual illusions.
• To explain brightness perception through a dynamic filtering process that manages edge representations.
• To demonstrate the model's ability to predict diverse visual illusions with consistent parameters.

Main Methods:

• A dynamic filtering process was developed to reduce redundancy and enhance non-redundant activity in edge representations.
• The dynamic filter was learned per input image, enabling context sensitivity.
• The model applied dynamic filtering to complex cell responses to create a gain control map, which then modulated simple cell responses for brightness mapping via activity propagation.

Main Results:

• The computational model successfully predicted numerous visual illusions, including contrast effects, assimilation, and reverse contrast.
• The model achieved these predictions using a single, consistent set of parameters across different illusions.
• The dynamic filtering approach demonstrated its efficacy in capturing context-dependent brightness perception.

Conclusions:

• Brightness perception can be explained by a dynamic filtering mechanism that processes spatial context.
• The proposed model offers a unified framework for understanding various visual illusions related to brightness.
• This approach provides insights into the neural computations underlying visual context sensitivity.