Research On Fingerppint Segmentation And Minutiae Matching Algorithm

Biometric identification technology is based on the identification of biological features of people. Due to its lifelong unchanged and uniqueness, fingerprint identification technology becomes the safest and most reliable identification technology. With a special conversion equipment and image processing technology, Automated Fingerprint Identification System (AFIS) can be quickly and accurately identify the individual identity through collection, analysis and comparison of fingerprints. AFIS has broad market prospects. Usually, AFIS consists of fingerprint collection, fingerprint preprocessing and fingerprint matching. Fingerprint preprocessing includes normalization, fingerprint segmentation, orientation computation, ridge distance estimation, fingerprint enhancement, fingerprint binarization, fingerprint thinning, and feature extraction. In this thesis, fingerprint matching and fingerprint segmentation have been studied. The main contents are introduced as follows.The segmentation method based on gray variance is simple and widely used, but it can hardly segment fingerprints with low contrast or high noise. A novel segmentation method based on ridge search is presented in this thesis. Two windows, which are the horizontal window and the vertical window are filled for each block through ridge search. Based on the prior knowledge, the new ridge coherence and the ridge distance variance of fingerprints are obtained for segmentation. The method we proposed can solve the problem effectively that segmentation results of fingerprints with low contrast or high noise are not satisfied using gray variance method. A hybrid method is also developed. Experiment results show that it achieves good segmentation results.Most existing minutiae based matching algorithms are grouped into the single reference point matching method. Among all possible point pairs, the pair of the most likely matched points are selected as the reference point pair. Due to the existence of nonlinear deformation of the fingerprint, positioning error of points farther away from the reference point pair is much larger which will lead to match error. Therefore, we raise multi-reference point pairs fusion method for minutiae based fingerprint matching. Firstly, several pairs of correct reference points are selected based on Consistency of Translation and Rotation Parameters (CTRP). Then every pair of reference points is used as the benchmark for the alignment of the template fingerprint and input fingerprint, so several set of minutiae pairs matched are acquired. Finally, we utilize information from sets to calculate match score. To some extent, the selection of multi-reference point pairs and the fusion of multiple set of minutiae pairs matched reduce the impact of nonlinear deformation on minutiae based fingerprint matching.The segmentation method based on ridge search makes up for the lack of the gray variance method. As it is a window based method, the segmentation results will be affected by the size of the window, and the segmentation result of regions where the fingerprint regions and the background regions come to junction are not satisfied. Therefore, the choice of window size and how to handle border area are the focuses of future research. The research on multi-reference point pairs fusion method for minutiae based fingerprint matching is still in the trial stage. Experiments verify the effectiveness of that select multi-reference points using TCRP. We utilized information of multiple set of minutiae pairs matched with the weighted fusion strategy to get match score. Experiments proved the feasibility of the matching algorithm proved to some extent, though match performance need to be further improved. In future, we focus on how to improve the accuracy of reference point pairs, reduce the time complexity of the algorithm, and choose more effective strategies to improve the match performance, etc.