Abstract
The methods for synthesizing facial images with arbitrary expressions on arbitrary viewpoints are applied to a lot of applications such as low bandwidth video conferencing. In this paper, our goal is to reconstruct Euclidean structure of a face with expressions from 2 uncalibrated real images using a knowledge for expression. What we have is the knowledge about the object category (faces in our case). Therefore, the recovered Euclidean structure is not necessarily a precise Euclidean description of that object, but is useful for many applications such as new image synthesis. By using this technique, Euclidean structure with expression was reconstructed and synthesizing new images achieved a fairly good performance.
