https://github.com/leaf918/Robust-Partial-Fingerprint-Alignment Abstracts. Real-world fingerprint matching is important for a fingerprint verification system for mobile devices. Most mobile and embedded devices have a small fingerprint capture sensor that can capture ONLY a portion of the fingerprint image. In this work, we present a fast and robust method for matching fingerprints with neural networks. We use a 2-point parameterization that maps the two corners of a fingerprint to another fingerprint. We use the SOCOFing dataset to train our network. The fingerprint alignment network works without local features extracted from the fingerprint images. There is a comparison between FP21Net and traditional homography estimation based on ORB features.