GAN-based Minutiae-driven Fingerprint Morphing

Fingerprint morphing is the process of combining two or more distinct fingerprints to create a new, morphed fingerprint that includes identity-related characteristics of all constituent fingerprints. Previously, this was done by either applying a model-based minutiaeoriented approach or a data-driven approach based on a Generative Adversarial Network (GAN). The model-based approach provides the ability to manage the number of minutiae coming from the fingerprints, but the resulting fingerprint often appears unrealistic. On the other hand, the data-driven approach produces realistic fingerprints, but it does not guarantee that the resulting fingerprint matches the original fingerprints. In this work, we introduce an algorithm that combines minutiae-oriented and GAN-based approaches to generate morphed fingerprints that look realistic and match their original fingerprints. The algorithm is initially designed to generate double-identity fingerprints and is further extended to generate triple-identity fingerprints. The results of our experiments indicate that the generated fingerprints appear realistic and the majority of them can be seen as double-identity fingerprints. The fingerprints resulting from morphing three fingerprints are unlikely to be triple-identity fingerprints, but rather anonymous ones matching none of the constituent original fingerprints.