基于主动形状模型的人脸特征建模

论文基于主动形状模型,提出了一种新的人脸特征建模方法,并详细介绍了唇部形状的建模过程。该方法从人脸数据库中提取一定的训练集,对唇部形状进行标注、校准和统计分析,可以得到唇部的变形模式,每种变形模式对应形状协方差矩阵的一个特征值,而最大的N(N<10)个特征值对应的主要变形模式可以还原出98%的形状变化。实验证明,该方法以非常小的信息损失为代价,仅用少量的参数就可以描述绝大部分的视觉特征。

Facial Feature Modeling Based on the Active Shape Model

This paper proposes a new method of modeling facial features based on the active shape model,and introduces the detailed procedure to model the lip contour.A set of training pictures of faces are first selected randomly from the face database.After labeling the contour,normalizing the shape and statistical analysis,the transformation modes of the lip can be gained corresponding to each eigenvalue of the shape's covariance matrix.The first main modes corresponding to the biggest eigenvalues can recover 98% of all the transformations.Experiments have proved that,using this method,a small quantity of parameters will be used to describe almost all the visual transformations,while very little visual information is lost.