Multilinear models for face synthesis
Multilinear models for face synthesis
复制标题
人脸合成的多线性模型
DOI：
--
复制
发表时间：
2004
期刊：
影响因子：
--
通讯作者：
J. Popović
中科院分区：
文献类型：
--
作者：
Daniel Vlasic;M. Brand;H. Pfister;J. Popović
多线性模型提供了一种自然的方式来模拟变异的异质来源。我们特别感兴趣的面部几何形状的变化，由于身份和表情的变化。在这种设置中，多线性模型能够捕获诸如微笑风格的特质，即微笑取决于身份参数。多线性模型的两个属性对动画师特别感兴趣：可分离性-表达可以变化，而身份保持不变，反之亦然;一致性-表达参数编码一个人的微笑将编码模型所涵盖的每个人的微笑，适合他们的面部几何形状和微笑风格。我们介绍的方法，使多线性模型动画的脸，解决两个关键障碍的实用工具。构建多线性模型的关键障碍是需要大量的数据（完全对应）来解释属性设置的每一种可能组合。使用多线性模型的关键问题是设计一个直观的控制界面。对于数据采集问题，我们展示了如何从一组不完整的高质量人脸扫描中估计一个详细的多线性模型。对于控制问题，我们展示了如何驱动这个模型与视频性能，提取身份，表情和姿态参数。
Multilinear models offer a natural way of modeling heterogenous sources of variation. We are specifically interested in facial geometry variations due to identity and expression changes. In this setting, the multilinear model is able to capture idiosyncrasies such as style of smiling, i.e. the smile depends on the identity parameters. Two properties of multilinear models are of particular interest to animators: Separability – expression can be varied while identity stays constant, and vice versa; and Consistency – expression parameters encoding a smile for one person will encode a smile for every person spanned by the model, appropriate to their facial geometry and style of smiling. We introduce methods that make multilinear models a practical tool for animating faces, addressing two key obstacles. The key obstacle in constructing a multilinear model is the vast amount of data (in full correspondence) needed to account for every possible combination of attribute settings. The key problem in using a multilinear model is devising an intuitive control interface. For the data-acquisition problem, we show how to estimate a detailed multilinear model from an incomplete set of high-quality face scans. For the control problem, we show how to drive this model with a video performance, extracting identity, expression, and pose parameters.