Study Of Feature Representation For Fingerprints And Its Applications

Fingerprint recognition system (AFIS) is one of the most robust biometric recognition methods. However, for large population or older people's poor-quality fingerprints, the conventional minutiae-based recognition method does not perform well, which has seriously constrain for many applications.In our opinion, one major reason is that the representation of the traditional technology is not sufficient. Using only minutiae can not describe all the information available in fingerprint images. Our motivation is to incorporate more features into fingerprint representation, and then develop novel recognition algorithms. Theoretically, this multiple-feature-based representation can strengthen the fingerprint individuality, which is the premise for using fingerprints as a biometric feature.Our works mainly lie in the following aspects. (1) We proposed two kinds of mathematical models for the orientation field, and developed an algorithm for orientation field extraction. It is proved that this algorithm is robust to the noises while requires few bytes to store the entire orientation field. (2) We proposed a novel algorithm for singular point detection using the global orientation field similarity and the topological analysis of fingerprint images. Compared with the conventional local-information-based methods, this algorithm can effectively removes the spurious singular points in the regions such as creases, scars, etc. (3) A fingerprint recognition algorithm by combining global information (orientation field) and local cues (minutiae) is proposed. Experimental results show that our algorithm outperforms the previous state-of-the-arts. Moreover, orientation field is used to facilitate fingerprint alignment, and proved to be helpful for poor-quality (such as older people) or incomplete (such latent images left on the spot) fingerprints.