摘要： 提出一种基于局部线性嵌入的最大散度矩阵算法——FSLLE。引入线性映射解决局部线性嵌入算法的样本外学习问题，通过自适应动态地确定局部线性空间邻域参数，最大化地融合样本数据的类别信息和局部结构信息矩阵，以获取髙维数据的最佳分类低维子空间。在JAFFE人脸表情库对该算法进行测试，结果表明，FSLLE算法能根据流形结构动态地确定局部邻域的大小，具有较好的表情识别率。

关键词: 局部线性嵌, 有监督学习, 表情识别, 流形学习, 最大散度矩阵

Abstract: A new expression image feature extraction and recognition method based on Locally Linear Embedding(LLE) and fisher criterion, named FSLLE, is proposed. The algorithm dynamically determines the local neighborhood size by using the relationship between its local estimated geodesic distance matrix and local Euclidean distance matrix and effectively integrated the face manifold local structure information with the labels’ information by adjustable factor. The proposed method is tested and evaluated using the JAFFE face expression database. Experimental results show that FSLEE validate the proposed approach and is more powerful for expression recognition.

Key words: Locally Linear Embedding(LLE), supervised learning, expression recognition, manifold learning, maximum scatter matrix

中图分类号: 

• TP391