Abstract: The similarity between facial modeling and its prototype is a key index in determining the success of the modeling. Traditional similarity research is based on isomorphic data, and there are few researches on the similarity between 2D face image and 3D facial mesh. This paper, based on two-layer Laplace manifold alignment, discovers the common manifold by dimensionality reduction on 2D face data set and 3D face data set with same number of samples, builds a similarity model between 2D face image and 3D facial mesh, and has a quantitative calculation on the similarity between isomerous faces. The validity of this method is testified by experiments.

Key words: 2D face, 3D face, similarity model, manifold alignment

摘要： 人脸艺术造型与其原型人脸的相似性是造型成功与否的关键指标之一。传统相似性研究建立在同构数据特征基础之上，对呈异构形态的二维图像人脸和三维网格人脸之间的相似性计算问题的研究还很少见。采用双层拉普拉斯流形对齐方法，通过对相同样本数的二维人脸数据集和三维人脸数据集进行协同降维，发现两者的共享流形嵌入，建立异构的二维人脸图像与三维网格人脸之间的相似模型，实现对异构人脸之间相似性的定量计算。通过实验，证明了该方法的合理性与有效性。

关键词: 二维人脸, 三维人脸, 相似模型, 流形对齐