Feature-Aware and Degradation-Driven Image Enhancement for Real-Time Fingerprint Recognition

The fingerprint sensor is widely used for biometric authentication in devices, such as smartphones and secure access control systems. However, environmental factors, such as low temperature and high humidity, degrade fingerprint image quality, impacting recognition accuracy. To address this, we propose a lightweight, feature-aware, and degradation-driven model named DMCRNet. Our approach introduces a two-stage architecture that incorporates a confidence map extractor to estimate ridge contours and minutiae points with high precision, followed by a degradation-driven network to enhance fingerprint quality while maintaining computational efficiency. DMCRNet improves equal error rate (EER) by 11.78% in FVC2004 DB1 and inference time by 97.42% in mobile devices compared to multistage training model [Cheng, et al (2023)], ensuring superior EERs with a compact 0.215 M size.