Optimized Deformed Residual Neural Network Accompanied by a Featured Discriminator for Wet and Fine-Grained Fingerprint Sensor Image

Fingerprint recognition is crucial for device and data

security, especially with the widespread use of capacitive sensors

in mobile devices. However, denoising wet fingerprints from

these sensors poses challenges due to small fingerprint areas,

limited features, and significant moisture-induced dark regions.

Our ”DRB-FD” method combines a Featured Discriminator (FD)

and a Deformed Residual Block (DRB) with attention mechanisms,

drop-out layers, and pre-activation. In experiments using the Nasic9395

0606 aug dataset , DRB-FD achieved a remarkable 73.1%

improvement, with the False Rejection Rate (FRR) decreasing by

25% compared to the baseline PGT-Net. Both FD and DRB significantly

contributed, showing improvements of 47.4% (FRR decreased

by 16.2%) and 48.9% (FRR decreased by 8.8%), respectively.

Besides achieving better FRR results, there is a significant increase in the number of fingerprint points successfully

matched after repair by the DRB-FD model.

In conclusion, our innovative approach effectively denoises wet fingerprints from capacitive sensors, enhancing the

accuracy and reliability of fingerprint recognition systems.