Graph Matching Algorithms Based On The Fingerprint Structural Features

In some current automatic fingerprint identification algorithms, there still exist severalproblems, such as, low orientation field accuracy, poor pre-processing effect, slow matching speed,poor recognition effect for low quality fingerprint images. In this paper, three new algorithms forthe fingerprint image pre-processing and the feature matching are proposed. To some extent, ouralgorithms not only enhance the speed and accuracy of fingerprint identification, but also raise therecognition rate of low-quality fingerprint images. The main attributions in this paper are asfollows:(1) This paper first introduces the pre-processing procedure of the fingerprint imageidentification system. The procedure mainly includes orientation field extraction, orientation fieldoptimization, enhancement, binarization and thinning. A new fingerprint image enhancementalgorithm based on the wavelet transform is proposed, and it can achieve a better performance.(2) Combining with an existing fingerprint quality evaluation system, an improved matchingalgorithm based on triangular topology is proposed. In the quality evaluation module of thisalgorithm, the quality of a fingerprint image is comprehensively evaluated by several factors, whichinclude the effective areas of fingerprint images, the wet and dry conditions，the offset degrees, thepresence or absence of singular points and the number of minutiae. According to the qualityevaluation results obtained from the fingerprint quality evaluation module, we adjust the matchingthreshold of the algorithm. The matching results of the low quality fingerprint images indicate thevalidity of the algorithm.(3) An improved matching algorithm based on the Delaunay Triangulation (DT) is proposed.In the algorithm, the local details and the global structure of the fingerprints are both considered.The specific procedures are as follows. First, according to the generated DT grids, we find thereference minutiae pairs and adjust the posture of the query fingerprint image. Then we use theminutiae-based algorithm to match the query image and the template image. Finally, we calculatethe similarity to determine the final match result. The result of the similarity is calculated by theminutiae direction, the type of the minutiae, the distance threshold and the matching number of theminutiae.The simulation results show that, comparing the two matching algorithms proposed in thepaper with the minutiae-based fingerprint matching algorithm proposed by Ren Qun and thefingerprint matching algorithm based on the Delaunay Triangulation (DT) proposed by Yin Yilong,the proposed algorithms can effectively reduce the false acceptance rate (FAR) and the false rejection rate (FRR).