Orientation Field Estimation for Latent Fingerprints by Exhaustive Search of Large Database

Abstract

Orientation field estimation is one of the most important steps in latent fingerprint recognition systems. However, due to very poor image quality, the performances of the state-of-the-art algorithms of orientation field estimation are still far from satisfactory. Based on the assumption that a sufficiently large fingerprint database should contain fingerprints whose orientation fields are quite similar to a latent fingerprint, we propose an orientation field estimation algorithm based on exhaustive search of large database for nearest neighbor. We set up a large database of 10,000 fingerprints of good quality, whose orientation fields and poses are estimated offline using traditional methods. Given a latent fingerprint as input, the most similar orientation field in the database is found and is combined with the original latent image to obtain the final orientation field. As a by product, we also obtain the pose of the latent fingerprint, which can be useful for fingerprint registration. The experimental results on NIST SD27 latent database show our method performs better than the state-of-the-art algorithms, in both orientation field estimation accuracy and identification performance, especially on those fingerprints of very poor quality.