基于二维图像三维人脸建模.doc

** ** ** ** 摘要 自Parke首次使用人脸参数模型生成三维人脸图像以来，构建逼真的三维人脸模型一直都是计算机视觉领域的一个研究热点。本文在对三维人脸建模技术认真分析的基础上，对基于二维图像的三维人脸建模方法进行了深入的研究。主要研究内容包括以下几个方面： (1) 从Poser7.0中导出通用三维人脸模型。根据MPEG-4的FDP标准，在正、侧面人脸照片上标出能基本反映人脸几何形状和结构特点的特征点。这两步为后面个性化人脸模型的创建奠定了基础。 (2) 采用了改进的径向基函数插值方法变换通用人脸网格模型,并对插值后的模型进行二次平滑处理，得到的特定人脸模型几何结构更加自然。 (3) 利用三张照片进行纹理合成。对三张照片分别进行匹配融合，然后对融合后的图像进行二次融合处理，使最终得到的纹理图像更加真实自然。根据合成的人脸纹理图像的特点，分别计算出人脸模型的各个部分的纹理坐标。 (4) 把模型数据和纹理数据导入到Visual C++6.0环境中利用OpenGL图形库进行纹理映射。 实验证明，本文提出的算法简便可行，有效避免了网格模型三角面片的扭曲，纹理映射效果生动逼真，得到了具有真实感的三维特定人脸模型。 关键词：三维人脸建模，特定人脸模型，径向基函数，纹理映射，OpenGL Abstract Since the F.I. Parke generated the first 3D human face image using a Parametric Model for Human Faces, it has been always a research hotspot to built vivid 3D human face model in the computer graphic area. After surveyed the 3D face modeling methods, the author carried out a thorough study in the 3D face modeling using 2D images. The research is composed as follows: (1). Exported the general face model from Poser 7.0. From front and side images, selected feature points, which correspond to MEPG-4 FDP. These feature points can usually reflect geometrical shape and characteristics of human face. These two steps lay the foundation for generating the upcoming individual fiscal model. (2). Deformed the generic face model using the improved Radical Basis Function Interpolation and applied smooth treatment basing on the post-interpolation face model. After this step, the geometrical shape of generated 3D face grid model becomes more natural. (3). Performed HYPERLINK /Periodical_jsjfzsjytxxxb200410016.aspx \t _blanktexture synthesis using three 2D images. Matched and smoothed these images separately. Then the texture images from the previous step was further performed second fusion, as a result, the human face texture image becomes much smoother. From this point, based on the characteristics of the synthesized face texture, every texture coordinates of the face model can be computed, respectively. (4). The model data and tex