Exploring Methods Used in Face Swapping

Abstract

Face swapping involves replacing the face in one image (the target) with a face in a different image (the source) while maintaining the pose and expression of the target face. Previous methods of face swapping required extensive computer power and man hours. As such, new methods are being developed that are quicker, less resource intensive, and more accessible to the non-expert. This paper provides background information on key methods used for face swapping and outlines three recently developed approaches: one based on generative adversarial networks, one based on linear 3D morphable models, and one based on encoder-decoders.