Marcel Grimmer defended his Master Thesis on Unknown Fingerprint Presentation Attack Detection Using Convolutional Autoencoders

Marcel Grimmer defended his Master Thesis on Unknown Fingerprint Presentation Attack Detection Using Convolutional Autoencoders

 

The increasing popularity of biometric authentication systems simultaneously raises privacy and security issues. Among other vulnerabilities, presentation attacks (PAs) that are directed to the capture device pose a severe threat. In order to be prepared against such attacks, presentation attack detection methods are deployed. However, due to the variety of materials that can be used to fabricate a presentation attack instrument (PAI), the classification models must be designed to also protect against unknown presentation attacks. The contribution of this thesis is the development of an unsupervised learning technique based on Convolutional Autoencoders and finger images stemming from two novel sensor technologies: Laser Speckle Contrast Imaging (LSCI) and Multi-Spectrum Short-Wave Infrared (SWIR). On an experimental evaluation over a database of 19, 598 bona fide images and 4, 226 PAs, including 43 unique PAIs, an average detection equal error rate of 2.47% could be achieved.

Marcel Grimmer defended his Master Thesis on Unknown Fingerprint Presentation Attack Detection Using Convolutional Autoencoders