Related Experiment Videos
Two-dimensional Gabor-type receptive field as derived by mutual information maximization
1NEC Corporation, Tsukuba, Japan
Summary
Simple cells in the visual cortex use Gabor functions to process visual information. This study demonstrates these receptive fields optimally extract maximum information from images, especially in noisy conditions.
Area of Science:
- Computational Neuroscience
- Information Theory
- Image Processing
Background:
- Simple cells in the visual cortex exhibit spatially localized, orientation- and spatial-frequency-tuned receptive fields.
- Gabor functions are widely accepted as accurate models for these receptive fields.
Purpose of the Study:
- To investigate the information-theoretic basis of two-dimensional receptive fields.
- To demonstrate that Gabor functions arise from maximizing mutual information.
- To suggest that simple cell receptive fields are optimally designed for information extraction.
Main Methods:
- Information-theoretic analysis
- Mutual information maximization framework
- Derivation of receptive field properties
Main Results:
- Gabor functions are derived as solutions to a mutual information maximization problem.
- In low signal-to-noise ratio conditions, Gabor-type receptive fields maximize information extraction from local image categories.
Conclusions:
- The findings support the hypothesis that simple cell receptive fields are optimally designed from an information-theoretic perspective.
- This provides a theoretical foundation for the prevalence and function of Gabor-like receptive fields in the visual cortex.