Realistic 2D Facial Animation from One Image

Abstract

In this paper we present a novel complete framework for creating realistic facial animation given only one neutral facial image as an input data. Our approach is carried on in a two-dimensional image space, instead of three-dimensional space. In addition, we employ an advanced computer vision method (digital image matting) as well as conventional image processing techniques (texture synthesis and image warping) in order to express more realistic facial animations. The major contribution of this work is showing how facial animation with a variety of realistic expressions can be generated very efficiently, where not only main facial components (e.g., eyeball, eyebrow and lip) but also pseudo-depth values obtained from their alpha mattes are utilized in our system. Simulations with real image confirm that our scheme produces high quality facial animations with an ease.