Brain responses strongly correlate with Weibull image statistics when processing natural images.

Area of Science:

• Neuroscience
• Computer Vision
• Image Processing

Background:

• Natural images possess a hidden statistical structure, often characterized by contrast distributions following a Weibull distribution.
• The parameters beta and gamma of the Weibull distribution are ecologically meaningful, relating to image fragmentation and texture.
• The brain is hypothesized to exploit these natural image statistics for efficient visual encoding.

Purpose of the Study:

• To investigate whether the human brain approximates the beta and gamma parameters of natural image contrast distributions.
• To explore the neural basis of processing these low-level image statistics.

Main Methods:

• A computational model was developed to estimate beta and gamma from simulated X-cell and Y-cell outputs.
• Electroencephalography (EEG) recorded brain activity (event-related potentials - ERPs) from subjects viewing natural images.
• The correlation between image beta/gamma values and EEG responses was analyzed.

Main Results:

• A model demonstrated that beta and gamma can be estimated from X-cell and Y-cell outputs.
• Beta and gamma values explained up to 71% of the variance in early ERP signals.
• These parameters outperformed other contrast measures in predicting neural responses.

Conclusions:

• The brain shows a strong tuning to the beta and gamma values of natural images.
• This tuning suggests an efficient neural mechanism for rapid image classification based on ubiquitous low-level statistics.
• Understanding these statistics may reveal fundamental principles of visual information processing.