Improving the performance of an HMM-based fingerprint recognition system

In this paper, general structure of a HMM-based fingerprint system is described. Since the HMMs have one-dimensional nature, in the presented system, Hilbert space-filling curves have been employed to map the fingerprint sample to a one-dimensional vector. The system have been simulated and tested on FVC2002-DB1 database with the EER rate of 4.2%. However, by investigating the samples that the system failed to recognize, it was concluded that by adding a sub-stage to the preprocessing, a portion of them can be detected. By applying this sub-stage, the EER rate were lowered to 2.85% at the cost of rejecting some samples.