用于表情克隆的基于混合边缘外观模型的面部动作追踪技术

FACIAL ACTIONS TRACKING FOR EXPRESSION CLONING BASED ON MIXTURE EDGE APPEARANCE MODELS

提出了基于混合观察模型的实时、全自动面部特征追踪技术.用一个三维参数化的模型用来为人脸和面部动作建模,同时用弱透视投影技术为头部姿态建模.WSF混合外观模型从基于非线性归一边缘强度的形状无关的局部片面被构建出来.实验结果表明,从观察模型和可适应混合学习中边缘强度的测量可以提高追踪的精确度和鲁棒性.

Facial feature tracking plays an important role in interactive entertainment, SUCh as expression cloning. Online-learning methods such as online appearance models have achieved good results in tracking, as they have strong abilities to adapt to variations. However, most previous works use only raw intensity to build observation models, which is very sensitive to illumination and expression changes. In this paper, a real time, fully automatic facial feature tracking approach using local structure based mixture observation model is presented. A 3D parameterized model is used to model face and facial actions, a weak perspective projection method is used to model head pose. WSF Mixture Appearance Models are built from shape-free patches based on non-linear normalized edge strength. Experimental results demonstrate that edge strength measures in observation modeling and adaptive mixture learning can improve accuracy and robustness of tracking.