How synthetic fingerprints can improve pre-selection of MCC pairs using local quality measures

A major source of errors in fingerprint recognition systems is poor quality of fingerprints. Local quality of fingerprints plays an important role in these systems to ensure high recognition performance. Recently an improved fingerprint matching method is proposed to use minutiae information encoded by Minutia Cylinder-Code (MCC) together with cylinder quality measures as local quality measures associated to each MCC descriptor. In this paper, we present our work where we have taken the advantage of a varying quality data set of synthetic fingerprint images in order to improve the pre-selection of MCC pairs using local quality measures. Since ground truth minutiae information is available for the synthetic fingerprints, we could create a large set of genuine/impostor minutiae as well as genuine/impostor MCC pairs. Subsequently a 2-class (genuine vs. impostor) classification model is proposed to modify the local similarity scores using two quality related local features, namely the cylinder quality measures and the number of extracted minutiae in the cylinders. Our experiments on synthetic and real data show that the local similarity scores modified through the proposed approach improve the pre-selection as well as global matching performance.