Project A7: Analyzing minutiae distributions for fingerprint recognition

There is a long tradition of using minutiae, i.e. endings and bifurcations of ridges (see picture), for fingerprint recognition. Typically, the positions and orientations of the minutiae are compared between fingerprints. It is imperative for realistic matching that minutiae quality be taken into account. Developing quality measures and their validation (improved quality relates to improved fingerprint matching in the use case) is currently a highly relevant topic in forensic application and research.

In the first cohort we have developed a new validation scheme and a novel method for quality estimation based on a wavelet-like image decomposition. Our new toolchain often outperforms state of the art quality measures and is currently being tested in collaboration with the federal bureau of criminal investigation. In the following cohort, further improving matching rates, we direct our attention to the relative importance of individual minutiae, using models from point process theory that are fitted and validated using Markov chain Monte Carlo and other Bayesian methods.

Methods: point processes, spatial statistics, Bayes classification, conformal geometry, Wasserstein metrics, non-Euclidean statistics
Applications: fingerprint analysis