Jesper Bang successfully defended his Master Thesis on „Ensuring Security in Biometrics – Presentation Attack Detection for Contactless Fingerprint Recognition“
In the current modern world, biometrics has become an integrated part of people’s daily lives. These systems provide fast and secure identification and verification for, e.g., mobile devices, removing the need for slower knowledge-based authentication. As the usage of biometric systems for recognition increase, so does the attacks against them. Therefore, Presentation Attack Detection methods are essential to help fight spoofing attempts and ensure system security. This work utilizes multiple versions of Convolutional Neural Networks and an autoencoder to identify attack presentations and seeks to improve these with hyperparameter tuning and added processing steps. Such steps involve obtaining relevant regions of interest from the images by masking and gaining information for classification using image Power Spectrum Energy (PSE). This work also proposes different fusions of presentation attack detection methods. These included combinations of some of the best performing versions of CNNs, the AE and the PSE. At a fixed Bona Fide Presentation Classification Error Rate (BPCER) of 0.2%, the fusions performed from 0% Attack Presentation Classification Error Rate (APCER) with Detection Equal Error Rate (D-EER) of 0% up to 10% APCER with 4% D-EER. The default performance of the autoencoder was 95.83% APCER on the contactless fingerprints. This work improved performance of the autoencoder significantly, and through model fusions even greater improvements were achieved. Most of the fusions were also suitable for a mobile scenario where an operator would be hand-holding the capture device.