Fingerprints Matching Using Spectral Graph Matching

Fingerprint matching plays a significant role in personal identification and recognition systems, that are mostly applicable in scenarios requiring high accuracy such as border crossings, crime suspects search engines, banking, e-commerce, etc.

Recent work focused on handling latent fingerprints, prints gathered in the wild. This work presents fingerprint matching system based on spectral assignment approach. The matching task relying mainly on minutiae points is modeled as graph matching problem approximately solved using spectral relaxation.

The proposed approach attains prominent results on challenging NIST-27 database, consisting latent-tenprint pairs with varying quality.

* M.Sc. research supervised by Prof. Yosi Keller