The subtle details on an expressional face,such as creases and furrows,are very important visual cues,but they are difficult to model and synthesize as they vary dynamically from one frame to another while people speak and make expression.A novel strategy,which is different from the traditional texture mapping methods,is proposed for generating such a kind of dynamic facial textures according to the motion of facial feature points.Based on the observation that shape and appearance on face images are highly correlated,a mapping is designed to transfer one image to the other.The mapping is called SADM(shape-appearance dependence mapping).The experimental results show that the synthesized faces with SADM are very close to the real ones.The proposed SADM strategy can be integrated into a wire-frame based head model to generate the realistic animation effects,or applied to a model-based video coding to produce more efficient bit-rates.%脸部表情的变化细节(例如皱纹)是很重要的视觉线索,但是它们难以建模与合成.这是由于人们说话和作表情时,脸部纹理也在动态地改变.与传统的纹理映射方法不同,提出一种根据脸部特征点运动来生成脸部动态纹理的新方法.基于形状和外观在表情脸部图像上高度相关这个观察,设计建立了从形状到外观的映射关系.这个映射被称作"形状外观的关联映射(shape-appearance dependence mapping,简称SADM)".实验结果表明用,SADM合成出的人脸与真实的人脸十分近似.提出的SADM方法可以集成到基于线框模型的人头模型中产生真实的动画效果,也可以应用于基于模型的视频编码来进一步节省传输带宽.
展开▼
