Based On The Reference Point Of The Structure Of The Fingerprint Minutiae Matching Algorithm

Biometrics identification is a kind of individual identification technology using the individual physiological characteristics. Fingerprints are the ideal basis for individual identification owning to their uniqueness and immutability. With the development of computer and information technology in recent years, as a mature application technique in the field of biometrics identification, fingerprint identification has been widely used in individual identity verification and authentication. Therefore, research on fingerprint identification has not only the important academic value, but also has a broad application prospect.The work reported in this thesis includes the technologies of fingerprint images enhancement and feature extraction, and the emphasis of the thesis focuses on the step of fingerprint matching. There exist many fingerprint matching algorithms at present. The minutiae-based matching algorithms have the longest history, and therefore the performance of them also has some comparative advantage. A kind of the most representative minutiae pattern matching algorithm—the minutiae feature string distance matching algorithm is introduced in this thesis. In this kind of algorithms, it is necessary to test all the possible minutiae pairs in order to extract the reference points, which leads a big computational cost. Considering the shortcoming and the high matching precision of structured methods, we can extract the reference point only in the local region of fingerprint images, and the structured methods can also be introduced into the stage of extracting reference point. Based on the above consideration, the reference point- based structured fingerprint minutiae feature matching algorithm is proposed in this thesis. Singular points (core and delta) are extracted firstly in the algorithm, and then we build a local structure for each minutia chosen around the fingerprint core. At last, the reference point can be extracted precisely by matching these local structures. In addition, a novel descriptive attribute of minutiae is also introduced into the matching step. In the matching stage, the minutiae pairs that are not matched obviously can be recognized rapidly through comparing the new attribute of them, and then the final decision is made based on the method of boundary box.In order to test the performance of the proposed algorithm, experiments are done on the four databases of FVC2000 under the standard of them. Comparisons of the results of the proposed algorithm and a minutiae feature string distance matching algorithm are also given in the thesis. Experimental results demonstrate that the average match times and the average enroll times of our algorithm on DB1-DB4 are less than the other one, and the matching precision is also basically better than the other one.