提出了一种基于图像重建的表情识别算法。首先，用LE（lipschitz embedding）算法提取出训练集中各个对象的表情流形，并建立流形向量与图像向量的映射关系。再通过非线性重建，确定待测图像在流形空间中的坐标。最后，用待测图像在各表情路径上的投影，重建各种表情图像，实现表情识别。该算法解决了各表情流形相互重叠的问题，且对表情强度变化具有鲁棒性。在Cohn-Kanade和CMU-AMP人脸库上的结果实验表明，该算法具有较好的表情识别率。

In this paper, an expression recognition algorithm based on image reconstruction is proposed. Firstly, manifolds of different subjects were obtained singly using the Lipschitz Embedding algorithm. Mapping between manifold vectors and image vectors was established. Then, the coordinates of testing image in manifold space was confirmed by nonlinear reconstructing. Finally, expression images were reconstructed according to projecting vectors of testing image on manifold paths and expression recognition was accomplished. The algorithm solves the problem that expression manifolds overlap with one another. It is robust to the variety of expression intensity. The experiments in Cohn-Kanade and CMU-AMP face database show the algorithm has better expression recognition accuracy.