3D deformable face model for pose determination and face synthesis

Abstract

This paper describes a scheme for pose determination and face synthesis using a deformable 3D face model. This model is a composite of three sub-models: edge model, color model and a wire frame model (WFM) which jointly describe the shape of the face and various facial features. The edge and color models are used for image analysis and the WFM is used mainly for face synthesis. We can project this 3D model onto different view planes based on three rotational parameters to generate 2D face templates for matching face images. After matching, the pose can be readily estimated. The composite face model is allowed to undergo complex deformation to handle face shape variations among different people. The WFM is incorporated into the model so that after matching, the WFM can be readily adapted to the image and image synthesis can be carried out. This technique can be used to analyze and synthesize face images over a wide range of poses.