CONTROLLABLE MANIPULATIONS OF FACIAL IMAGES

Abstract

This thesis targets with controllable manipulations of facial images. Specifically, it first proposes a method for synthesis of facial images with virtual identity to protect the face template of users in face verification systems. It then addresses the simultaneous manipulations of pose and illumination for facial images. Particularly, it proposes a multi-attribute facial image generator that is can be learned from disjoint partially-attributed datasets.