On the relationship between optical variability, visual saliency, and eye fixations: a computational approach

Area of Science:

• Visual perception
• Computational neuroscience
• Image processing

Background:

• Current models of visual saliency lack a unified framework linking physical properties to perception.
• The efficient coding hypothesis provides a foundation for understanding visual system efficiency.
• Contextual adaptation mechanisms in the visual system aim to maintain consistent behavioral responses.

Purpose of the Study:

• To propose a hierarchical definition of optical variability grounded in the efficient coding hypothesis.
• To investigate the role of contextual adaptation in ensuring behavioral invariance.
• To test the proposed framework against human eye-tracking data and novel hyperspectral imaging experiments.

Main Methods:

• Developed a hierarchical definition of optical variability connecting physical magnitudes to visual saliency.
• Compared the proposed model with human fixation data across three eye-tracking datasets, including face-specific images.
• Utilized hyperspectral representations of surface reflectance in a novel experiment.
• Evaluated the model's ability to explain a visual illusion without eye movements.

Main Results:

• The proposed definition offers a more reductionist interpretation of optical variability compared to existing methods.
• Results on face images showed a significant decrease in the influence of semantic factors.
• Hyperspectral image analysis supported the assumptions for estimating optical variability.
• The model successfully explained quantitative findings related to a visual illusion.

Conclusions:

• The hierarchical definition of optical variability provides a robust and parsimonious framework for understanding visual perception.
• Contextual adaptation mechanisms are crucial for maintaining stable behavioral responses to visual input.
• The findings suggest a reduced role for semantic influences in saliency, particularly in complex image datasets.