近似人脸全视角图生成及在特定人脸模型映射

Near Global Texture Generating and Its Mapping on Specific Face Model

用二维人脸照片构造三维特定人脸模型是三维人脸建模研究的热点问题,其中三维人脸纹理映射直接影响着生成三维人脸的逼真度。采用基于人脸正侧面照片特征点的分层、分部位的径向基函数重构一般人脸模型,得到部位轮廓特征明显的三维特定人脸模型,并提出了基于拉普拉斯金字塔分解的拼接图像交接处色彩融合算法,对人脸正侧面照片进行拼接融合,得到近似的人脸全视角纹理图。然后将近似人脸全视角纹理图映射到三维特定人脸模型,从而得到三维特定真实感人脸模型。实验结果表明该方法计算近似人脸全视角纹理图简单,生成特定人脸模型效果逼真,可应用于特定人脸动画制作的逼真人脸建模中。

Constructing 3-D specific face model with 2-D face photographs is an active topic in 3-D face modeling research, in which texture mapping of 3-D face directly affects the fidelity of 3-D face reconstructed. The method of multi-layer and multi-region with RBFs based on feature points of the front and side face photographs was adopted for reconstructing the generic face model to get 3-D specific face model with visible organ character. A method of color-fusion on the boundary of mosaic images based on Laplacian Pyramid transform was proposed, which could merge and fuse front and side face photographs and obtain the near face global texture. Then the near face global texture was mapped onto the reconstructed specific face model and the 3-D realistic specific human face model was gotten, Experiments show that this method is easy for constructing near face global texture and the reconstructed specific face model is natural and realistic. The method can be used for realistic specific face modeling in facial animation.