A Multiple Substructure Matching Algorithm for Fingerprint Verification

In this paper, we represent a fingerprint image with a Delaunay graph formed by minutiae as nodes. The graph has attributes which contributes to the final similarity measure and are invariant under rotation and translation. We design an algorithm for the comparison of these graphs based on the similarities of multiple common substructures. We use different heuristics to tackle the problems of noise, deformation and partial matching found in fingerprint recognition. We match star structures and extend it by edges maintaining the local structural compatibility. Finally, we consolidate the global similarity taking into account the size of the common substructures and the accumulated similarity of all stars involved. We use a simple greedy algorithm obtaining a very efficient performance. We use our proposed method in some experiments with fingerprint images in databases from FVC2002. It shows better results compared to other known algorithms as K-plet, and several others recently published.