Towards online iris and periocular recognition under relaxed imaging constraints

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Accurate iris segmentation is crucial for reliable iris recognition, especially with distantly acquired images.
• Existing methods struggle with less imaging-constrained environments, necessitating robust segmentation and recognition strategies.

Purpose of the Study:

• To develop an efficient iris segmentation and recognition approach for distantly acquired images.
• To improve the accuracy and robustness of iris recognition systems in unconstrained environments.

Main Methods:

• A novel iris segmentation approach utilizing a random walker algorithm for initial segmentation.
• Post-processing techniques to refine segmentation accuracy.
• Integration of simultaneously extracted periocular features with iris features for enhanced recognition.

Main Results:

• Achieved significant improvements in average segmentation accuracy across UBIRIS.v2 (9.5%), FRGC (4.3%), and CASIA.v4-distance (25.7%) databases.
• Demonstrated substantial gains in recognition performance by combining iris and periocular features, with average improvements of 132.3%, 7.45%, and 17.5% on the respective databases.

Conclusions:

• The proposed iris segmentation method offers superior accuracy compared to state-of-the-art algorithms.
• Joint segmentation and feature combination strategies significantly enhance iris recognition performance in challenging conditions.