A Flexible Object Model for Recognising and Synthesising Facial Expressions

We here introduce the Flexible Object Model to represent objects with structured deformation, such as the human face under variable expression. The model represents object shape and texture separately and extracts a data parameterisation autonomously from image sequences after initialisation by a single hand-labeled model graph. We apply the model to the representation, recognition and reconstruction of nine different facial expressions. After training, the model is capable of automatically finding facial landmarks, extracting deformation parameters and reconstructing faces in any of the learned expressions.