图像局部保留投影及在表情识别中的应用

Image locality preserving projection and its application to expression recognition

提出了一种基于图像局部保留投影(ILPP)的人脸表情识别方法,该算法在降维过程中将图像结构信息融入到局部保留投影(LPP)目标函数中。与LPP不同,ILPP通过降维处理,获得表情图像结构信息的同时将投影最优化,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CED-WYU(1.0)和JAFFE两个表情数据库的识别结果表明,ILPP特征提取方法能有效地提高识别率。

A facial expression recognition method based on Image Locality Preserving Projection（ILPP） was proposed.Unlike Locality Preserving Projection（LPP）,ILPP incorporated graph construction into the LPP objective function in dimension reduction process,and thus obtained a simultaneous learning framework for graph construction and projection optimization,ILPP could extract more useful and decisional expression features from the original expression data.The experimental results on CED-WYU（1.0） and JAFFE show that ILPP is an effective method for improving the recognition accuracy.