基于面部特征约束的人脸纹理映射及变形  

Face texture mapping and deformation with constrains of facial feature

作　　者：罗岱[1] 陶洋 杨刚[1] 

机构地区：[1]北京林业大学,北京100083

出　　处：《计算机工程与应用》2018年第6期188-192,240,共6页Computer Engineering and Applications

基　　金：国家科技支撑计划课题(No.2015BAH52F03)

摘　　要：将一幅人脸图像(称为源图像)映射到另一幅人脸图像(称为目标图像)上,从而实现人脸图像的变形并达到某种特殊效果的表现要求。其中,一个最关键的问题是保持两幅人脸中器官特征的约束映射。提出一种技术策略,可以有效实现这种映射效果。利用人脸特征检测算法检测出源人脸和目标人脸的特征点;基于特征点,利用Delaunay三角剖分的方法将目标人脸转化为三角网格;采用调和映射的方法计算得出所有网格点在源人脸图像中的对应坐标;在获得网格点纹理坐标后,就可以实现源人脸到目标人脸的映射。实验表明,该方法可以实现任意人脸间的映射,并能较好地保持人脸五官间的映射关系。The research object of this paper is to map a face image（called the source image）to another face image（called the target image）, so as to realize the deformation of the face images and achieve some special effects. In the process, one of the most critical problems is to maintain the constrained mapping of organ features in two human faces. This paper puts forward a technical strategy, which can effectively achieve this object. Firstly, the feature points of the source face and the target face are detected with the face feature detection algorithm. Then, based on the feature points, the Delaunay triangulation method is used to transform the target face into a triangular mesh. After that, the corresponding coordinates of all grid points are calculated in the source face image by using harmonic mapping. After obtaining the texture coordinates of the mesh point, the mapping of the source face to the target face can be realized. Experimental results show that, the proposed method can realize the mapping between arbitrary faces, and the mapping relationships of facial features can be maintained very well.

关 键 词：纹理映射 调和映射 DELAUNAY三角化 ASM算法 

分 类 号：TN911.9[电子电信—通信与信息系统]