Iris Recognition Method Based on Parallel Iris Localization Algorithm and Deep Learning Iris Verification.

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Biometric recognition, particularly iris recognition, is crucial for security but struggles with noisy, unnormalized images from non-cooperative environments.
• Current deep learning methods for iris recognition often increase model complexity and neglect the essential iris localization stage.

Purpose of the Study:

• To develop an effective and efficient iris recognition scheme addressing challenges in non-cooperative environments.
• To improve iris localization accuracy and speed.
• To create a lightweight deep learning model for high-precision iris verification with reduced complexity.

Main Methods:

• Iris localization using parallel Hough circle transform and the Daugman algorithm.
• Iris verification employing a novel lightweight convolutional neural network with deep residual network modules and residual pooling.
• Experiments conducted on multiple iris datasets collected under non-cooperative conditions.

Main Results:

• Significantly accelerated iris localization speed (21-36x) on GPU compared to CPU across four datasets.
• Achieved effective iris localization accuracy.
• Demonstrated high-precision iris verification with a lightweight network, yielding low equal error rates (1.01%-1.71%) across four databases.

Conclusions:

• The proposed scheme effectively addresses iris recognition challenges in non-cooperative environments.
• The novel lightweight CNN offers a balance of high accuracy and reduced model complexity for iris verification.
• This approach enhances the practicality and efficiency of iris recognition systems for security applications.