An information maximization model of eye movements

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Vision Science

Background:

• Human eye movements are crucial for visual perception.
• Current models often rely on saliency or random processes to predict fixation points.
• Understanding the principles guiding eye movement programming is essential for vision research.

Purpose of the Study:

• To propose a novel computational model for programming eye movements.
• To investigate information maximization as a core principle for guiding visual attention.
• To compare the predictive accuracy of this model against existing approaches.

Main Methods:

• Developed a sequential information maximization model for eye movements.
• Reconstructed high-resolution visual information from fixation sequences.
• Accounted for the decrease in visual resolution from the fovea to the periphery.
• Compared model predictions with human eye movement data and saliency/random models.

Main Results:

• The information maximization model significantly outperformed saliency and random models in predicting human fixation locations.
• The model's core principle is to select the next fixation point that minimizes uncertainty about the visual stimulus.
• The framework successfully reconstructs visual information across a sequence of fixations.

Conclusions:

• Information maximization provides a powerful and effective principle for programming eye movements.
• The proposed model offers a more accurate prediction of fixation sequences compared to existing methods.
• Incorporating additional biological constraints could further enhance the model's predictive capabilities.