Cross-sensor iris recognition through kernel learning.

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Iris biometrics are increasingly popular, leading to frequent sensor upgrades and deployments.
• Re-enrollment for new sensors is costly and time-consuming for large user bases.
• Cross-sensor matching in iris recognition often results in significant performance degradation.

Purpose of the Study:

• To develop a machine learning technique to mitigate performance loss in cross-sensor iris recognition.
• To adapt iris samples from one sensor to another, enabling seamless recognition across different devices.
• To improve the accuracy and efficiency of iris recognition systems without requiring re-enrollment.

Main Methods:

• A novel optimization framework for learning transformations on iris biometrics was developed.
• This framework was used for sensor adaptation by minimizing intra-class distance and maximizing inter-class distance.
• The method ensures class separability irrespective of the sensor used for image acquisition.

Main Results:

• Extensive evaluations on multi-sensor iris data demonstrated improved cross-sensor recognition accuracy.
• The proposed method effectively reduces the performance gap between same-sensor and cross-sensor matching.
• The technique shows significant improvements in iris recognition performance across diverse sensor types.

Conclusions:

• The proposed machine learning technique successfully addresses cross-sensor performance degradation in iris recognition.
• The method offers a practical solution for integrating new sensors without compromising existing biometric data.
• Minimal integration effort allows easy adoption into current iris recognition pipelines, enhancing system robustness.