Computational modeling of visual salience alteration and its application to eye-movement data

Area of Science:

• Computational neuroscience
• Visual attention modeling
• Artificial intelligence

Background:

• Computational saliency map models are crucial for understanding how visual salience influences attention.
• Existing models primarily focus on either functional approximation or neurobiological implementation.
• A gap exists in models that integrate both functional performance and neurobiological plausibility.

Purpose of the Study:

• To develop a novel computational saliency map model integrating functional approximation and neurobiological implementation.
• To evaluate the model's predictive performance against conventional methods using eye-movement data.
• To explore the model's capability for simulating neural-level effects on visual salience.

Main Methods:

• Proposed a novel saliency map model incorporating diverse image features and center-surround competition via an artificial neural network.
• Evaluated the model's predictive accuracy using an open eye-movement dataset.
• Conducted neural-level simulations by altering parameters related to neural activity and synaptic density.

Main Results:

• The novel model demonstrated predictive performance comparable to conventional saliency map models for eye-movement analysis.
• Simulations revealed that changes in excitatory-inhibitory balance, baseline activity, and synaptic density modulated salience weighting.
• The model successfully quantified changes in salience weighting reflected in eye movements.

Conclusions:

• The integrated model effectively bridges predictive and neurobiological perspectives in visual salience research.
• This novel approach offers a powerful tool for investigating the mechanisms of abnormal visual salience.
• The findings pave the way for new strategies in understanding and potentially treating visual attention disorders.