Abstract: Extracting facial expression features accurately and designing the superior performance classifier are the two critical problems in the field of facial expression recognition, and the recognition is more challenge for the East Asian expression who are famous for implicit. In order to solve this problem, the Two-Dimensional Linear Discriminate Locality Preserving Analysis（2D-LDLPA） algorithm based on image based projection is firstly proposed to directly extract features on image matrix. Then, through constructing the decision trees to learn the extracted expression features, the random forest classifier is applied to improve the accurate ratio of the East Asian facial expression recognition. Finally, the recognition accuracy and computational efficiency of the proposed algorithm are verified by the experiments on two East Asian facial expression databases.

Key words: East Asian facial expression recognition, manifold learning, Two-Dimensional Linear Discriminate Locality Preserving Analysis（2D-LDLPA） feature extraction, random forest classification

摘要： 如何准确提取出人脸的表情特征并设计出性能优越的分类器是表情识别研究中最关键的两个问题，尤其对于以含蓄著称的东亚人表情，识别起来更加困难。针对这一难点，首先基于图像投影的方式，提出二维线性判别保局分析（2D-LDLPA），直接在表情图像的二维矩阵中提取特征，然后，以提高东亚人脸表情识别率为目标，通过构造决策树来学习提取出的表情特征，进而使用随机森林算法实现分类，最后，在两个东亚人脸表情库上测试，验证所提算法识别的准确性和计算效率。

关键词: 东亚人脸表情识别, 流形学习, 二维线性判别保局分析（2D-LDLPA）特征提取, 随机森林分类