University of Twente Student Theses


Use of autoencoders for fingerprint encoding and comparison

Pool, W.R. (2021) Use of autoencoders for fingerprint encoding and comparison.

[img] PDF
Abstract:Deep learning methods for fingerprint comparison are starting to outperform classic, minutiae-based fingerprint comparison methods. This research explored the possibility of deep learning fingerprint comparison in two steps. First, an autoencoder was used to create a latent vector representation of a fingerprint segment. Three different autoencoders were created, one with a loss function which focuses more on fingerprint minutiae, one which emphasizes frequency components that normally occur less often in autoencoder reconstruction and one which is trained with normal mean square error. Secondly, a fully connected neural network was used to perform fingerprint comparison using the latent vectors of the autoencoders. The fingerprint comparison results show that despite some limitations in the training process and fingerprint data, it is possible to compare fingerprints using the latent space. It also shows that the classification results will improve with the different loss functions implemented in this paper. The most effective method was increasing the loss of the autoencoder at the minutiae locations.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:53 electrotechnology
Programme:Electrical Engineering MSc (60353)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page