Analysis of geometric and latent attributes in morphed face images

Supervisor(s)Dr. Cecilia Pasquini


Face morph attacks combine characteristics of multiple distinct subjects into one face image. This causes face verification systems to match the morphed face with all involved subjects. This thesis should first analyze geometric attributes of morphed face images and then explore the merits of synthesizing morphs with generative adversarial networks. The evaluation should compare the visual appearance as well as the detectability of the resulting morphs with state-of-the-art face morph detectors.