【摘要】 给出了基于实拍人脸图像的三维逼真人脸模型的重建算法 .该算法首先在两幅人脸图像上交互标识特征点对和输入摄像机的广角参数来实现摄像机定标 ,进而匹配出两幅人脸图像上的其它对应点 ,实现模型的三维重建 .作者用半自动方法来达到匹配目的 :用手工编辑建立的二维对应网格 ,得到初始人脸外形 ;采用鲁棒的最大似然立体匹配算法自动匹配出稠密的对应点 ,重建出表示人脸的散乱三维数据点团 ;最后利用这些稠密的三维数据点去迭代矫正和自适应细分手工编辑的三维初始人脸网格得到结果模型 .鉴于所得网格往往存在噪声 ,该算法利用 L aplace二阶算子对它进行光滑处理 ,并给出了纹理映射的绘制结果 .实验结果表明 ,该算法无需昂贵的设备 ,既省力 ,又有很强的真实感效果更多还原

【Abstract】 This paper presents a reconstruction algorithm for 3D realistic facial model based on facial images. In this algorithm, the camera is calibrated according to its field of view and manual labeled feature points. The dense corresponding points on the two facial images are matched using a semi _automatic method to carry out 3D reconstruction. Firstly, the initial face shape is gained by editing 2D corresponding mesh manually. Secondly, a robust maximum likelihood stereo matching algorithm is used to get dense corresponding points, and recover their 3D positions. Lastly, the 3D initial facial mesh is subdivided adaptively according to the distribution density of 3D points to construct the resulting model. To remove the reconstruct noise, second order Laplace operator is used to smooth the model. The result model is decorated by texture mapping. Experiments show that the algorithm is simple and can generate very realistic facial expression without expensive devices.更多还原