Deep Residual CNN-Based Ocular Recognition Based on Rough Pupil Detection in the Images by NIR Camera Sensor

Area of Science:

• Biometrics and Pattern Recognition
• Computer Vision
• Artificial Intelligence

Background:

• Accurate iris segmentation is critical for iris recognition, but low-quality images (blurring, obstructions) reduce accuracy.
• Deep learning improves segmentation but is computationally intensive, requiring long processing times.
• Existing methods struggle with image quality issues and misalignment, impacting recognition rates.

Purpose of the Study:

• To develop a novel, efficient iris recognition method that overcomes limitations of traditional segmentation.
• To enhance recognition accuracy and speed, particularly for challenging iris images.
• To propose a new preprocessing and recognition strategy for robust biometric identification.

Main Methods:

• Developed a rapid method to identify a rough ocular area, bypassing precise iris segmentation.
• Utilized the broader ocular region for recognition to mitigate issues from iris segmentation inaccuracies.
• Employed a deep residual network (ResNet) to address misalignment between enrolled and recognition images.
• Validated the method on three diverse iris databases: CASIA-Iris-Distance, CASIA-Iris-Lamp, and CASIA-Iris-Thousand.

Main Results:

• The proposed method demonstrated higher recognition accuracy compared to existing techniques.
• Achieved improved performance despite image quality challenges like blurring and reflections.
• Effectively handled misalignment issues through the use of ResNet and ocular area recognition.
• Showcased significant improvements across multiple benchmark iris datasets.

Conclusions:

• The novel ocular area recognition method offers a faster and more accurate alternative to traditional iris segmentation for iris recognition.
• This approach provides a robust solution for iris recognition systems dealing with low-quality images and potential misalignment.
• The study highlights the potential of deep residual networks and broader ocular region analysis for advancing biometric security.