Abstract
Fingerprints have been widely used as personal identification for many years. Although there are several biometric techniques recently, fingerprint still has its own advantage due to large databases which may be established for many years. However, personal identification in such a large database is difficult and time consuming. In this thesis, we implement an automatic fingerprint identification system (AFIS) with these stages: image pre-processing, minutiae extraction, fingerprint classification and fingerprint matching. In the first stage, Gabor filter is used to enhance the furrow and to reduce the noise on the fingerprint image. Then, minutiae (ridge endings and bifurcations) are detected for matching. The stage of fingerprint classification is to reduce the matching time. A fingerprint is classified into one of the five types: Arch, Right Loop, Left Loop, Whorl and Others. In the final stage, a matching score is computed by comparing minutiae patterns between two fingerprint images. The AFIS is tested on 6 databases of fingerprint images, such as Rindex28, Lindex101 from PRIP Lab at NTHU and DB1, DB2, DB3, DB4 provided by FVC2000, on a PC with Pentium 4 3.00GHz CPU and 1 GB SDRAM running Windows XP. Based on the criterion of reaching top 3 of the matching scores, the identification rates are 100% (112/112), 93.32% (377/404), 97.50% (78/80), 92.50% (74/80), 86.25(69/80) and 92.50% (74/80) by testing the aforementioned fingerprint image databases, respectively.