Dhanesh Budhrani defended his Master Thesis on „R-MFID: Reference Morphed Face Image Detection“

The application of biometric systems for access control scenarios allow us to authenticate with our own biological or behavioural characteristics. Different biometric modalities such as the face, iris, fingerprint, etc. present diverse ad- vantages and disadvantages in specific scenarios. For instance, the face was chosen in 2002 by the International Civil Aviation Organization (ICAO) as the primary biometric modality for verification in the electronic Machine Readable Travel Document (eMRTD), due to the convenient access to this biometric characteristic. This ease of access allows to compare the stored information of the face in the eMRTD with a live capture of the subject in an Automated Border Control (ABC). However, a study presented in 2014 proved the vulnerability of ABC scenarios against morphed face image attacks. Such attack consists of the intro- duction into the eMRTD of a synthetic face image that blends the face images of two (or more) different subjects. When such a morphed sample is enrolled, it allows the authentication of more than one subject against the face image stored in the eMRTD. This vulnerability implies a serious threat to civil secu- rity, since this attack could be used by criminals to cross an ABC gate and, for this reason, extensive research is being conducted over the last 3 years by nu- merous research groups. There are only two published research papers to date that propose a system to detect such synthetic artifacts, both analyzing micro-textures of the suspicious image (i.e., no-reference detection systems). These systems, however, have not proved to generalize well to datasets on which they have not been trained. The goal of this thesis is to create the first detection system of morphed face images that makes use of a bona fide image (captured at the ABC gate) in order to compare it to the image stored in the eMRTD (i.e., reference detection system), and determine whether the latter image is morphed or genuine.