摘要
针对真实感表情合成中的难点——人类表情的种类多样性和个体特性,提出一种分层人脸模型.该模型将人脸图像分为基础层、运动层和细节层3层,基础层采用主动外观模型来表示表情运动带来的全局性变化,对于运动层提出一种基于隐马尔可夫模型的运动纹理分块组合机制来实现风格化运动纹理,细节层通过几何变形来恢复人脸个体特征.实验结果表明,文中提出的运动纹理分块组合策略有效地避免了样本表情种类不足的问题,能够组合出样本以外的风格表情.此外,该模型简单有效,能同时合成出表情种类的多样性和个体特性.
A multi-layer facial model is proposed to resolve a challenging problem in photorealistic facial expression synthesis--the diversity and individuality of expressions. The model decomposes the face image into three layers： the base layer, motion layer and residue layer. In base layer, the active appearance model is used to represent global and general changes of expressions. In motion layer, a motion texture subdivision and combination mechanism is proposed based on the hidden Markov model to generate new kinds of motion texture, which is the key design to model expression＇s diversity and individuality. In residue layer, wrapping technique is used to retrieve and synthesize facial texture details. Experimental results demonstrate that the proposed motion texture subdivision and combination mechanism efficiently avoid the insufficiency of sample＇s expression types and generate individual expressions. Additionally, the proposed facial layered model is simple and useful, and is able to synthesize diversiform and individual facial expressions.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2013年第8期1204-1212,共9页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61175029)
