A Novel Probabilistic Model Based Fingerprint Recognition Algorithm

Abstract

A novel fingerprint recognition algorithm based on the probabilistic graphical model is proposed in this paper. First, minutiae in query fingerprint are viewed as random variables with the minutiae in template print as the realizations. According to the random variables, a 2-tree model is built by selecting two signal points from the query set. Second, the model is converted into a Junction Tree, and the potentials of the tree nodes are defined according to the intrinsic characters of fingerprint. After that, Junction Tree (J.T.) algorithm is performed to obtain the correspondence of the two sets of minutiae. To deal with many-to-one corresponding problem caused by the outliers, we repeat the process by exchanging two sets. Finally, the similarity of the two fingerprints is evaluated using the number of common matching pairs and the maximal posteriori probability generated by the J.T. algorithm. Experiments performed on databases of FVC2004 achieve the perfect performance.