Stacking Fingerprint Matching Algorithms for Latent Fingerprint Identification

摘要

Automatic latent fingerprint identification is still challenging for biometric researchers. One infrequently explored approach for improving the identification rate involves stacking latent fingerprint identification algorithms with a supervised classification algorithm, instead of using a weighted sum or a product of likelihood ratio. A stacking approach fuses the result provided by different base algorithms to achieve higher performance than each individual algorithm. Latent fingerprints present different qualities, causing deviations between the identification rates of various algorithms. Thus, we propose stacking latent fingerprint identification algorithms using a supervised classifier. We use two different minutia descriptors with a global matching algorithm independent of the local matching of the minutia descriptor. Our stacking method improves the identification rate of each base algorithm by 2% when comparing the fingerprints in the database NIST SD27. Furthermore, our proposal achieves a 73.26% rank-1 identification rate when comparing 258 samples in the database NIST SD27 against 29,258 references, and 68.99% against 100,000 references.