Application of convolutional neural networks for fingerprint recognition

Abstract: Fingerprint recognition is a well-known problem in pattern recognition and widely used in contemporary authentication technology such as access devices in mobile phones. The subject of this thesis is to investigate the applicability of convolutional neural networks for fingerprint recognition. This is accomplished by designing various network architectures for this task. Our starting-point is an architecture known as a siamese network, from which we build upon by including additional components as well as network architectures based on the siamese architecture. The networks are realized by implementation. Data for training and evaluating the networks is provided as gray-scale images of fingerprints and we implement a simple algorithm for generating ground truth labels. To evaluate our work, we measure the performance of all implemented models with common metrics for fingerprint recognition algorithms. Lastly problems with our approach are listed and potential future improvements are given.