A hierarchical registration algorithm for fingerprints from multi-type capture sensors

Various types of fingerprint sensors introduce large variances of fingerprints in distortion patterns and noise, which greatly challenge the traditional matching algorithms. In this paper, we develop a hierarchical registration algorithm to handle the non-linear distortion among multiple kinds of sensors. The transformation model is initially estimated with the traditional rigid model and gradually upgraded to affine and quadratic model. The model upgradation is carefully designed to obtain the trade-off between the fitting accuracy of high-level model and the facilities of low-level model. We finally establish the minutiae correspondence with a modified ICP (Iterative Closest Point) method. Experimental results demonstrate that our algorithm effectively improves the performance of cross-matching, especially for those databases in different modes of acquisition.