Abstract: To solve the problem of blendshapes generation mostly depending on manual modeling and fine-tuning,this paper proposes an automatic generation method about example-based expression blendshapes. Using a set of generic face models as the prior,it makes each generated blendshapes approach to its real semantics. Expression blendshapes generated by this method can map semantics and expression dynamics from the generic model to the target model. The method achieves model’s scalability in application and enables users to gradually calculate more expression models according to their own needs. Experimental results show that the method,constructing a gradient space to optimize blendshapes, outperforms the state-of-the-art methods both in speed and reality.

Key words: blendshape expression radical, example-based, iterative optimization, gradient space, reality, facial animation

摘要： 针对现有blendshape 表情基的生成多依赖手工建模和微调的问题,提出一种基于样例的表情基自动生成方法。以一组通用的面部模型为先验模型,实现每个blendshape 的表情语义逐步求精。通过该方法生成的blendshape 表情基能够将通用模型的表情语义和表情动态映射到目标模型上,从而实现模型在应用上的可扩展性,使用户能够按照自身需求逐步计算出更为丰富的面部表情模型。实验结果表明,在梯度空间建立形变约束,并将其转化为一个关于blendshape 的优化问题,能更加有效地控制blendshape 表情基的变化趋势,并且计算时间更短,真实感更强。

关键词: blendshape 表情基, 基于样例, 迭代优化, 梯度空间, 真实感, 表情动画

CLC Number: 

• TP391