Robustness of Deep Learning Based Face Recognition Under Morphing Attacks

Last decades with the development of deep learning techniques the evident advances have been reached in the field of face recognition. However at the same time more evolved and sophisticated techniques for performing the presentation attacks continue to appear, which require the development of new protection solutions.

One of such face image manipulating methods is Face Morphing. Image morphing techniques are used to combine information from two (or more) images into one image. Over the past decade, it has gained significant attention and has been more thoroughly investigated. As awareness of the problem has grown, numerous counterfeit documents employing face morphing techniques have been uncovered at control gates.

Given the importance of reducing vulnerabilities in modern face recognition systems and the significant risks posed by presentation at tacks, this thesis aims to contribute with the tools for combating the face morphing problem involving deep learning algorithms.

Best Doctoral Consortium Award