单视影像下的人脸快速三维重建

3D Face Reconstruction for Single Visual Images

提出了一种单视影像下的人脸快速三维重建方案,能够对非可控条件下获取的单幅人脸图像进行有效、快速的三维重建。在三维人脸重建阶段,提出了一种特征动态选择策略,针对待重建人脸图像动态选择合适的2D-3D特征变换子空间,在保证执行效率的同时能显著提高三维人脸模型的重建精度。实验证明,本方法不仅简洁高效,而且对人脸姿态和光照变化具较好的鲁棒性,在WHU-3DFaceData数据库上取得了理想的重建效果。

This paper proposes a fast and efficient method of 3D face reconstruction for single visual images which are obtained under uncontrolled conditions. First of all, face feature points are located through the ASMS （active shape model of subset）. During this locating processing, ASMS locating accuracy is improved by increasing the completeness and the clus- tering performance of the training samples, which also sets up the fundamental basis for fur-ther face feature extraction. Secondly, 2D-3D feature subspaces are built up according to dif- ferent subsets respectively using method LFA （local feature analysis）, which efficiently fu- ses the global and local features. Thirdly, a dynamic selection strategy for 2D-3D features subspace selection is designed, which can significantly improve the reconstruction accuracy and keep a content processing efficiency at the same time. What is more, Experimental re- sults show that the proposed method is not only simple and efficient, but also robust to pose and illumination variations, and achieves desired reconstruction results on WHU-3DFaceData database.