Two-step subspace learning for texture synthesis of facial images

In recent years, animation of dynamic facial expression got many attentions in entertainment and other fields. In this paper, we describe a new useful method for synthesis of facial images (ex. different expression, different view point) from one natural facial image. A lot of methods such as subspace learning have been proposed for synthesis of facial images. But the synthesis accuracy by existing methods is not enough, especially for texture synthesis. In this paper, we propose a two-step subspace learning method to improve the synthesis accuracy. In our proposed method, we add a residual error subspace learning step for reduction of synthesis error. The proposed method has been applied to synthesize expressional facial images.