基于二维图像的三维人脸建模(2)

根据二维图像重建三维图形

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 texture 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 texture data are imported into Visual C++ 6.0 to perform texture mapping using OpenGL graphic library. This algorithm has been proved easier and feasible by the experiment. It can effectively prevent triangular facets of the grid model from distortion, making the texture mapping effect more vivid and giving specific 3D face model more realistic.

Key words: 3D face modeling, specific face model, Radial Basis Function, texture mapping, OpenGL

II