Periocular Data Fusion for Age and Gender Classification

Area of Science:

• Biometrics and Security
• Computer Vision
• Machine Learning

Background:

• Soft biometrics are gaining traction in security and business for individual identification.
• Single soft biometric traits have limitations; combining multiple sources improves accuracy.
• The periocular region offers a viable alternative for biometrics when lower facial data is unavailable.

Purpose of the Study:

• To investigate the fusion of periocular features (pupils, fixations, blinks) for demographic classification (age and gender).
• To evaluate different data fusion strategies at both feature and score levels.

Main Methods:

• Feature-level fusion using a concatenation scheme.
• Score-level fusion employing transformation and classifier-based methods (weighted sum, weighted product, Bayesian rule).
• Utilized periocular data including pupils, fixations, and blinks for classification.

Main Results:

• Achieved 84.45% accuracy for age classification.
• Achieved 84.62% accuracy for gender classification.
• Demonstrated the effectiveness of data fusion for enhancing soft biometric performance.

Conclusions:

• Fusion of periocular soft biometrics is effective for demographic classification.
• The proposed method balances privacy protection with strong discriminatory power.
• Encourages further research into periocular biometrics for applications where lower facial data is obscured.