重建发型的真实感头部建模

Realistic Human Head Modeling with Hairstyle Reconstruction

提出了从几何模型和多视点彩色图像构造真实感三维头部模型的方法.作为输入数据的彩色图像不仅被用来构造全景纹理图,而且是发型重构的数据源.首先在模型空间定义对应于真实图像平面的虚拟图像平面,恢复不同视点图像中发型外轮廓点的相应三维位置;然后根据发根位置、生长方向和长度,在模型的头皮表面重建发型.另一种发型重构的方法是从多幅图像中抽取的二维发型轮廓线出发,通过立体匹配法恢复其三维位置,然后用Coons-patch法重构发型表面.实验结果显示了文中方法的头部重建效果.

We present methods for generating photo-realistic 3D model and multi-view color images of a person. The input color human head models from a geometric images are used not only to generate panoramic texture maps, but also as the source data for hairstyle reconstruction. At first, we define the virtual photo plane, corresponding to the true photo plane, to recover 3D positions of hair contour points extracted from the multi-view color images. Then, by using the growth root, direction and length of every hair, hairs are grown till covering the scalp. Another hair surface reconstruction approach starts with extracting contour points of hairs from input images and applying the stereo-vision method to recover their 3D positions. Then, a Coons-patch is constructed to approximate the hair surface from recovered four closed 3D contour curves. The experimental results show the performance of the methods for different real models and images.