U-Net-Based Fingerprint Enhancement for 3D Fingerprint Recognition

Area of Science:

• Biometrics and Security Engineering
• Computer Vision
• Machine Learning

Background:

• Conventional 2D fingerprint biometrics face sensor spoofing and disease transmission risks due to contact-based sensors.
• Three-dimensional (3D) fingerprint recognition offers contactless capture and enhanced security against spoofing.
• Converting 3D fingerprint data (point clouds) to 2D images often results in poor contrast, hindering recognition accuracy.

Purpose of the Study:

• To develop an effective image segmentation approach for enhancing the contrast of 3D fingerprint images.
• To improve the performance of conventional 2D fingerprint recognition methods when applied to 3D fingerprint data.
• To address the limitations of existing 3D to 2D fingerprint conversion processes.

Main Methods:

• Proposed an image segmentation method utilizing the deep learning U-Net architecture.
• Applied the U-Net model to enhance the contrast of 2D gray-scale images generated from 3D fingerprint point clouds.
• Evaluated the enhanced fingerprint images using conventional fingerprint recognition techniques.

Main Results:

• The proposed U-Net based image segmentation significantly improved fingerprint contrast.
• Fingerprint recognition Equal Error Rate (EER) decreased from 41.32% to 13.96% in experiment A.
• EER improved from 41.97% to 12.49% in experiment B, demonstrating substantial performance gains.

Conclusions:

• Deep learning-based image segmentation effectively enhances 3D fingerprint images for recognition.
• The proposed method offers a viable solution to improve the accuracy and reliability of 3D fingerprint authentication systems.
• This approach addresses key challenges in 3D fingerprint recognition, paving the way for more secure and hygienic biometric solutions.