Residual Finger Feature Points UNet with Nonlinear Decay on Wet Partial Fingerprint Image Recognition in Tiny Sensor

Abstract—Fingerprint recognition technology has becomepopular for mobile device authentication systems due to itsreliability and ease of use. As smartphones evolve, fingerprintsensors are now integrated into smartphone power with a widthof 2.2 mm. However, tiny sensor sizes have led to limited fingercoverage and external factors such as sweat or water dropletscan cause image distortion, making user authentication morechallenging.To address these issues, we propose the FFP-UNet, which usesfingerprint feature points-based restoration and nonlinear decayrate residual within a U-shape architecture to recognize blurryand wet fingerprints within size constraints effectively. Ourapproach aims to restore fingerprints effectively while avoidingover-restoration and preserving local matching feature points,which reduces the false rejection rate (FRR). We develop novelresidual blocks, which feature a nonlinear decay mechanismthat optimizes weight allocation between blocks to enhancefeature extraction capabilities. Furthermore, our residual featurepoints fusion module restores contextual information by fusingmatching feature points and features from the previous level. Toovercome unaligned data in real-world scenarios, our approachincorporates the residual MaxBlurPool module.Through comprehensive experiments on real-world data, weachieved a remarkable 9.4% reduction in FRR. Our methodoutperforms the basic U-Net framework [1] by 50.52% inoverall performance. We improved 48.35% over FPDMNet [2]and 58.77% over DenseUNet [3] which is trained with smallpatches. Moreover, PGT-Net [4] is also used for small areasof wet fingerprints, but our FFP-UNet outperformed it by72.51%. Source code is available for research purposes athttps://github.com/aannn555/FFP-UNet.