Abstract: Fingerprint matching may include one or more of the following techniques. Space-frequency representations are adaptively computed for one or more fingerprint images. Key feature points of the fingerprint images are automatically extracted. Mesoscopic features are extracted based on the key feature points and the space-frequency representations. Fingerprint images are matched against a database of known fingerprint images using a matching algorithm based on the key points and mesoscopic features of the fingerprint images. Deep neural networks may be used for some or all of these steps.