具有照片真实感的三维特定人脸的生成

目前人脸建模的方法都不是很令人满意：成本高昂，算法复杂。针对这一困难提出了一种简单有效的由正、侧面照片重建特定人脸三维模型的新方法。先用多分辨率技术构造纹理无缝拼接图，然后把三维网格模型投影到圆柱面上并展开，根据二维图像变形技术中特征指定算法的数学思想，将匹配公式进行一定的改进后应用于模型网格点的适配，在作了一定假设后推导逆向圆柱映射的计算公式，将适配后的柱面网格映射回去，最后进行纹理映射生成真实感的特定人脸三维模型。实验证明，该方法可实现不同人脸的真实再现，而且无需昂贵的设备，算法简单，可以廉价的生成具有照片真实感的三维特定人脸。

On Inexpensive but Photo-Realistic 3D Modeling of Human Faces

Existing methods for modeling human faces are quite good except that the expense is high and that the algorithms are complex. An inexpensive, relatively simple but effective method is presented in this paper. In the full paper, we explain our method in much detail. Here we give only a briefing of our explanation. First we use multiresolution to construct seamless texture map, and unwrap the cylindrical projection of a general model mesh. Then warp and fit it to the texture map based on the improved equation of 2D image metamorphosis. After that we present a formula of inverse cylindrical projection to construct 3D specific face model with texture mapping. We used our method to model one human face. The results show preliminarily that our method does give photo-realistic 3D modeling. Compared with existing methods, our method's precision is a little lower; this price is worth paying when the decrease in cost and the simplification of algorithm are considered.