Unsupervised eye pupil localization through differential geometry and local self-similarity matching

Area of Science:

• Computer Vision
• Biometrics
• Human-Computer Interaction

Background:

• Accurate eye center localization is crucial for applications like gaze estimation and neurological screening.
• Current methods often rely on invasive devices or supervised machine learning, limiting their practicality and requiring expert setup.
• Existing unsupervised methods struggle with non-ideal imaging conditions and user variability.

Purpose of the Study:

• To develop a novel unsupervised approach for automatic eye center detection.
• To address the limitations of supervised methods in terms of adaptability and ease of use.
• To improve the accuracy and robustness of eye tracking in challenging scenarios.

Main Methods:

• The proposed method utilizes a novel representation of eye shape derived from differential image intensity analysis.
• Self-similarity coefficients are combined with shape information to capture local appearance variability.
• The approach operates on periocular patches, accommodating variations in scale, rotation, and translation.

Main Results:

• The unsupervised method demonstrated effectiveness on challenging facial image databases.
• Experimental results show favorable comparisons with leading state-of-the-art supervised methods.
• The approach achieves accurate eye center localization without requiring initial training sessions.

Conclusions:

• The new unsupervised method offers a promising alternative for automatic eye center detection.
• Its robustness to non-ideal conditions and lack of need for training make it highly versatile.
• This advancement has significant implications for accessible and adaptable eye-tracking technologies.