Morphing attacks on face recognition systems constitute a critical threat that in-
creases along with the technological developments in Machine Learning, especially
with Generative Adversarial Networks (GAN) and diffusion models. Advanced im-
age generation models are able to create images using morphing techniques that
can bypass the detection security systems, underlining the need for ways to ad-
dress the issue. This thesis focuses on the role of diffusion models in morphing
attacks and their attack detection.
During the course of this thesis, various models such as Inception V3, VGG-16
and Resnet-50 were evaluated, measuring their accuracy in attack detection while
trained on public datasets such as Face Recognition Grand Challenge (FRGC) and
Face Research Lab London (FRLL) dataset. The experiments showcase that the
morphing attack potential of well used Face Recognition Systems (FRS) can reach
over 90%. The MixFaceNet-Morphing Attack Detection (MAD) model, used on
the evaluation of SMDD (Synthetic-Face-Morphing-Attack-Detection-Development)
dataset has obtained an accuracy of 38% while detecting the attacks. The ex-
periments show the limitations of models in generalizing to unknown datasets
suggesting further research.