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

Biometrics-based authentication mechanisms can address the built-in weakness of conventional password or token-based authentication in identifying genuine users. However, 2D-based fingerprint biometrics authentication faces the problem of sensor spoofing attacks. In addition, most 2D fingerprint sensors are contact-based, which can boost the spread of deadly diseases such as the COVID-19 virus. Three-dimensional fingerprint-based recognition is the emerging technology that can effectively address the above issues. A 3D fingerprint is captured contactlessly and can be represented by a 3D point cloud, which is strong against sensor spoofing attacks. To apply conventional 2D fingerprint recognition methods to 3D fingerprints, the 3D point cloud needs to be converted into a 2D gray-scale image. However, the contrast of the generated image is often not of good quality for direct matching. In this work, we propose an image segmentation approach using the deep learning U-Net to enhance the fingerprint contrast. The enhanced fingerprint images are then used for conventional fingerprint recognition. By applying the proposed method, the fingerprint recognition Equal Error Rate (EER) in experiment A and B improved from 41.32% and 41.97% to 13.96 and 12.49%, respectively, over the public dataset.