人脸表情的形变线性拟合方法  

Morphable Linear Fitting Method for Facial Expression Synthesis

作　　者：李旭东[1] 张振跃[1] 

机构地区：[1]浙江大学数学系

出　　处：《自动化学报》2008年第5期593-597,共5页Acta Automatica Sinica

基　　金：国家自然科学基金(60372033);浙江省自然科学基金(Y107039);浙江省教育厅科研计划项目(20060121)资助~~

摘　　要：提出了用于人脸表情合成的形变线性拟合方法.该方法利用人脸图像形变模型线性组合逼近的基本思想,确定合成表情图像的形状信息和纹理信息,其步骤简单,容易实现.该方法能有效地从中性表情人脸图像合成出具有表情的图像,并且得到的人脸表情自然、逼真、具有说服力.更为重要的是,该方法能从闭着嘴的中性表情人脸图像合成出具有张开嘴露出牙齿效果的人脸表情图像,克服了当前大多数人脸表情合成方法不能实现这一效果的不足.A method of morphable linear fitting for facial expression synthesis is proposed. This method first gains shape vectors and texture vectors from given sample facial images of the same kind of expression. Then, based on the facial image morphable model, the linear combinations of the shape vectors and texture vectors of the unexpressive images are retrieved respectively using optimal linear fitting. Those combination vectors are kept in synthesizing the expressive image required. This method can be realized easily and the resulting image produced by this method is quite natural, lively and convincing. More importantly, this method can synthesize a facial image with mouth opened and teeth showed from an unexpressive image with closed mouth. It outperforms other existing methods for facial expression synthesizing.

关 键 词：人脸图像合成 人脸表情合成 形变模型 表情动画 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]