摘要
采用运动恢复结构(structure from motion,SFM)算法进行三维人脸建模一直以来受到研究者的关注,但其对错误的匹配点比较敏感,因此,文章提出了一种融合Gabor特征的SFM算法三维人脸建模方法。该方法利用Gabor滤波器提取纹理特征,判别轮廓特征点匹配的准确性;针对图像数增多,传统因子分解法不易修正旋转矩阵的问题,利用旋转矩阵的性质求得修正矩阵,避开方程组的求解;提出引入迭代最近点算法将稀疏三维特征点与三维模型进行配准,缩小空间距离,并结合薄板样条函数插值生成特定的三维人脸模型,为增强真实感,进行纹理映射。实验结果表明,该方法有效提高了匹配点的准确性,能够重建出具有较强真实感的三维人脸。
3D face modeling based on structure from motion（SFM） has always attracted researchers＇ attention. But it is sensitive to false matching feature points. In this paper, a method of fusing Gabor features for 3D face modeling by SFM algorithm is presented. In this method, texture features extrac- ted by Gabor filter are introduced to evaluate the reliability of outline feature points. With more ima- ges, the traditional factorization method is not easy to correct rotation matrix. To solve this problem, the property of a rotation matrix is adopted to calculate corrective matrix. It also avoids solving the e- quations. Then, sparse 3D feature points and the 3D model are registered by using the iterative closest point（ICP） algorithm to reduce the space distance. And combining the modified 3D model and the shape obtained by thin plate spline interpolation, the person-specific 3D face model can be reconstruc- ted. Finally, the texture is mapped to enhance the realism. The experimental results show that the proposed method is effective in improving the accuracy of matching points and reconstructing higher realistic 3D faces.
出处
《合肥工业大学学报（自然科学版）》
CAS
北大核心
2017年第2期180-185,共6页
Journal of Hefei University of Technology：Natural Science
基金
国家自然科学基金资助项目(61371156)
安徽省科技攻关计划资助项目(1401B042019)
