A Non-Rigid Registration Method with Application to Distorted Fingerprint Matching
Sheng Lan,Zhenhua Guo,Jane You
DOI: https://doi.org/10.1016/j.patcog.2019.05.021
IF: 8
2019-01-01
Pattern Recognition
Abstract:Fingerprints are unique and invariant, so they are widely used for biometric recognition. However, due to the problem of deformation in the actual sampling process, it may cause a great change in the fingerprint features. As the accuracy of fingerprint recognition depends on the quality of fingerprints, non-rigid registration algorithms are particularly important. Most existing non-rigid registration algorithms estimate the distortion only by minutiae and gratitude of fingerprints, while direction of the ridges is neglected or used indirectly. In this paper, we proposed a novel model based algorithm for non-rigid fingerprint registration using image fields. As direction information is very important for spatial transformation in registration and the image fields contain the direction of fingerprint ridges, by combing image fields with the traditional model based algorithm, we directly introduce orientation of the ridges to the model for estimating the distortion, and thus make a better use of the direction information of fingerprints and simplify the deformation model. Considering that delta/cores and minutia are sometimes hard to extract accurately and reliably, we used the whole image for matching directly. Experiments have been carried out on four representative databases, namely FVC2004 DB1, Tsinghua Distorted Fingerprint database, NIST SD27 database and NIST SD30 database. We also compared our algorithm with other state-of-the-art algorithms, the experimental results show the effectiveness of the proposed algorithm.