Abstract

Fingerprint recognition research based on Level 3 features – especially sweat pores – has got increasing interest thanks to its ability to operate under daunting conditions, such as matching latent and partial prints. In this work, we review methods, datasets, and training and evaluation protocols for pore detection intended for obtaining such features. We have observed many inconsistencies in training and evaluation protocols, data withholding, and lack of public source code have hampered reproducibility and comparisons in the literature. We aim to address these challenges by looking into the most promising insights from existing works to establish best practices and introduce a more reasonable starting point for future research. To do so, we create a baseline pore detector and reimplement three others for comparison purposes. We carried out our experiments using the most popular dataset – PolyU-HRF – and two recent publicly available datasets – L3-SF and IITI-HRF. Our results show a reproducible path for researchers and highlight that there is still a wide margin for innovation and improvement in this area. An open repository containing the source code for our self-implemented detectors and the protocols employed in our experimental evaluation is available in: https://github.com/azimIbragimov/Fingerprint-Pore-Detection-A-Survey