Latent Fingerprint Matching: Fusion of Manually Marked and Derived Minutiae

Matching unknown latent fingerprints lifted from crime scenes to full (rolled or plain) fingerprints in law enforcement databases is of critical importance for combating crime and fighting terrorism. Compared to good quality full fingerprints acquired using live-scan or inking methods during enrollment, latent fingerprints are often smudgy and blurred, capture only a small finger area, and have large nonlinear distortion. For this reason, features (minutiae and singular points) in latents are typically manually marked by trained latent examiners. However, this introduces an undesired interoperability problem between latent examiners and automatic fingerprint identification systems (AFIS); the features marked by examiners are not always compatible with those automatically extracted by AFIS, resulting in reduced matching accuracy. While the use of automatically extracted minutiae from latents can avoid interoperability problem, such minutiae tend to be very unreliable, because of the poor quality of latents. In this paper, we improve latent to full fingerprint matching accuracy by combining manually marked (ground truth) minutiae with automatically extracted minutiae. Experimental results on a public domain database, NIST SD27, demonstrate the effectiveness of the proposed algorithm.