Towards Efficient 3D Facial Appearance Models

Abstract

We present a novel approach to building realistic 3D facial appearance models. We seek to efficiently model temporal dynamics of a single human face by encoding raw 3D geometrical data and texture from scanners in a model suitable for statistical treatment. Preparation of our models requires minimal human interaction and, once built, our models can be used for real-time facial animation intuitively controlled by a small number of parameters, ready for rendering on consumer-class graphics hardware. Our method robustly handles poor-quality or deficient sample meshes and allows for processing of shape data in 2D mode using established image processing algorithms.