The method of vicinity minutiae decomposition with higher level graphs for fingerprint verification

Abstract

The subject matter of the paper is the development of fingerprint local structures based on the new method of the minutia vicinity decomposition (MVD) for the solution to the task of fingerprint verification. It is an essential task because it is produced attempts to introduce biometric technology in different areas of social and state life: criminology, access control system, mobile device applications, banking. The goal is to develop real number vectors that can respond to criteria for biometric template protection schemes such as irreversibility with the corresponding accuracy of equal error rate (EER). The problem to be solved is the problem of accuracy in the case of verification because there are false minutiae, disappearing of truth minutiae and there are also linear and angular deformations. The method is the new method of MVD that used the level of graphs with many a point from 7 to 3. This scheme of decomposition is shown in this paper; such a variant of decomposition is never used in science articles. The following results were obtained: description of a new method for fingerprint verification. The new metric for creating vectors of real numbers were suggested – a minimal path for points in the graphs. Also, the algorithm for finding out minimal paths for points was proposed in the graphs because the classic algorithm has a problem in some cases with many points being 6. These problems are crossing and excluding arcs are in the path. The way of sorting out such problems was suggested and examples are given for several points are 20. Results of false rejection rate (FRR), false acceptance rate (FAR), EER are shown in the paper. In this paper, the level of EER is 33 % with full search. 78400 false and 1400 true tests were conducted. The method does not use such metrics as distances and angles, which are used in the classical method of MVD and will be used in future papers. This result is shown for total coincidences of real number, not a similarity that it is used at verifications. It is a good result in this case because the result from the method index-of-max is 40 %.