嵌入保局投影及在表情识别中的应用

摘要

提出了一种基于嵌入保局投影的人脸表情识别方法,称之为ELPP.通过降维处理,ELPP不仅保留邻空间的局部结构信息,通过样本化的图像嵌入处理,还保留了嵌入空间的数据信息,这样可从原始表情数据中提取更多更有效更具判决性的表情特征信息.基于JAFFE和CED-WYU两个表情数据库的识别结果表明,基于ELPP的特征提取方法能有效地改善识别效果.%A facial expression recognition method based on Embedded Locality Preserving Projections(ELPP) is proposed. Unlike LPP, ELPP incorporates sampled graph construction in dimension reduction process, thus preserves the local manifold structure embedded in the ambient space. Whereas, ELPP aims still to preserve structures in data into the embedding space. ELPP can extract more useful and discriminatal expression features from the orginal expression data. Experimental result on JAFFE and CED-WYU shows that ELPP is an effective method for improving the recognition accuracy.