Nowadays, technologies of automatic identity authentication become more and more important. While so many drawbacks in traditional methods, such as keys or passwords, people turn to biometrics, a new research field with bright future. Fingerprint recognition, one of the most important branches in biometrics and the oldest identity authentication method, is showing with its unique advantages. Thousands of valuable results gained after over 20 years of research in this field. However, there are still difficult problems, such as enhancement of poor quality images, feature selection of fingerprint images and the matching of highly distorted fingerprint images. Survey on recent research of elastic fingerprint matching shows that these methods focus only on the representation ability of all kinds of distortion, but none of them question the validity of the registration in a whole. The main contribution of this thesis includes: 1. Survey of all kinds of fingerprint feature representation schemes. A proposed scheme of representation that facilitates the matching of distorted fingerprint images. 2. Introduce the concept of "registration pattern" in the first time. And proposed a novel elastic fingerprint matching method which optimally register all parts of the two fingerprints and check the validity of the "registration pattern" before performing the final matching. Knowledge of "genuine registration patterns" was gained by computing patterns from simulated distorted match by matching images extracted from NIST 24. "lest on the FVC2002 2nd database show that this algorithm has very good accuracy of fingerprint verification.
修改评论