Abstract: In order to restrain LLE’s sensibility to noise，a novel method combining Haar wavelet transformation is presented for face recognition.First，the original image data is decomposed into the high frequency components and the low frequency components by the Haar wavelet transformation.The horizontal and vertical high frequency components are ignored，and the original image data splice with its low frequency components by row stacking.Secondly，with the original image data reduced dimension by LLE，there are two matrices that correspond respectively with all samples between the training set and the testing set which are consisting of the original image data.Finally，the face classification is accomplished by the nearest neighbor criteria.The experimental results in ORL face-database and Sheffield face-database show that the proposed method is efficient to improve the recognition rate of LLE.

Key words: face recognition, manifold, Locally Linear Embedding（LLE）, Haar wavelet transformation, nearest neighbor classification

摘要： 为了抑制局部线性嵌入算法对噪音的敏感性，结合Haar小波变换，提出了一种人脸识别的新方法。利用Haar小波变换将原始图像数据分解为高频分量和低频分量，忽略水平高频与垂直高频分量，并将低频分量按行堆叠的方式引入其原始图像数据中。通过LLE对该图像数据进行降维，求得训练和测试样本各自对应的矩阵。依据最近邻准则完成人脸识别。基于ORL与Sheffield人脸数据库的实验结果表明了该方法对改善传统LLE算法识别率的有效性。

关键词: 人脸识别, 流形, 局部线性嵌入, Haar小波变换, 最近邻分类