Face Synthesis with Arbitrary Pose and Expression from Several Images - An Integration of Image-Based and Model-Based Approaches

We propose a method for synthesizing face views with arbitrary poses and expressions combining multiple face images. In this method, arbitrary views of a 3-D face can be generated without explicit reconstruction of its 3-D shape. The 2-D coordinate values of a set of feature points in an arbitrary facial pose and expression can be represented as a linear combination of those in the input images. Face images are synthesized by mapping the blended texture from multiple input images. By using the input images which have the actual facial expressions, realistic face views can be synthesized.