CNN-based Pore Detection and Description for High-Resolution Fingerprint Recognition

High-resolution fingerprint recognition usually relies on sophisticated matching algorithms to match hand-crafted keypoint, usually pores, descriptors. In this work, we improve the state-of-the-art results in a public benchmark by using instead a CNN pore descriptor with a simpler matching algorithm. We describe how aligning images allows learning keypoint descriptors when the dataset does not provide keypoint identity annotations. We also propose a new pore detector, together with a detailed evaluation protocol for this task. All our code is available at https://github.com/gdahia/high-res-fingerprint-recognition