Partial Fingerprint Matching via Feature Similarity and Pre-training

Existing partial fingerprint matching methods use fingerprint ridge features and minutiae or employ algorithms like SIFT and A-KAZA to create new feature points that can replace minutiae for feature extraction and matching. While these methods have achieved some success in improving matching performance, they rely on manually designed rules for extracting local area features, which limits their accuracy and generalization capability. To address these limitations, this paper proposes a novel partial fingerprint matching algorithm that leverages Feature Similarity and Pre-training. Specifically, Feature Similarity is integrated into a deep learning-based model to emulate traditional partial fingerprint matching techniques. Additionally, Pre-training guides the model to learn subtle yet identity-discriminative features within partial fingerprints. Experimental results on partial fingerprint databases constructed from FVC2004 DB1, DB2, and DB3 show that our algorithm achieves low EER and ZeroFMR, outperforming several state-of-the-art matching methods.