Cross-Sensor Fingerprint Enhancement Using Adversarial Learning and Edge Loss.

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Fingerprint sensor interoperability, or cross-sensor matching, is a challenge due to variations in noise and artifacts from different sensor technologies.
• Existing methods struggle to effectively enhance fingerprints captured across diverse sensor types.

Purpose of the Study:

• To develop a novel algorithm for enhancing fingerprint images acquired from different sensors and touch technologies.
• To address the limitations of current fingerprint enhancement techniques in cross-sensor scenarios.

Main Methods:

• Formulated fingerprint enhancement as an image-to-image transformation task using a deep encoder-decoder model.
• Employed both conventional and adversarial learning frameworks, including a conditional Generative Adversarial Network (cGAN).
• Incorporated edge loss during training, inspired by the ridge patterns in fingerprints, and evaluated using MOLF and FingerPass datasets.

Main Results:

• Achieved effective fingerprint quality enhancement across various sensor types, an area previously under-investigated.
• Demonstrated superior performance compared to state-of-the-art methods in fingerprint enhancement.
• Validated results using standard metrics like NBIS Fingerprint Image Quality (NFIQ) and Structural Similarity Index Metric (SSIM), alongside a proposed Fingerprint Quality Enhancement Index (FQEI).

Conclusions:

• The proposed deep learning-based method effectively enhances fingerprint images, overcoming cross-sensor interoperability challenges.
• The algorithm offers a significant advancement in fingerprint recognition systems by improving image quality regardless of the enrollment sensor.
• This work provides a robust solution for improving the reliability of fingerprint matching in diverse real-world applications.