一种基于全部和局部平滑光流的分步人脸重建算法

Step by step 3D face reconstruction algorithm based on CLG optical flow

提出一种基于全部和局部(combined local-global approach,CLG)平滑光流的分步人脸重构算法。首先利用原始CLG得到图像初始匹配值;再以反投影残差、光流梯度、人脸光流范围进行约束,找出初始匹配中不可信匹配区域;最后对不可信图像区域提取纹理并再次计算光流值。实验表明,该人脸重建算法的重建精度与鲁棒性比原始CLG算法高,能够得到光顺的人脸三维数据。

This paper proposed a step by step 3D face reconstruction algorithm based on CLG optical flow. First, it accomplished original image matching points using CLG algorithm. Then, it found uncertainty matching areas in original matching points automatically based on the constraint of reprojected error, optical flow gradient and face optical flow range. Last, it recomputed the optical flow value of uncertainty matching areas with the texture part of face images as input images. Real experimental results show that both the reconstruction accuracy and robustness with algorithm in this paper are better than with the original CLG algorithm. Smooth and vivid 3D face data can be got with the proposed algorithm.