摘要： 运用保局投影(LPP)算法进行人脸识别时，噪声会破坏真实流形。为此，提出一种解决噪声的新方法——HaarLPP方法。该方法利用Haar小波变换降低噪声的影响，运用LPP算法进行降维，依据最近邻准则完成人脸识别。基于AT&T与Sheffiled人脸数据库的实验结果表明，该方法在噪声的敏感性方面优于传统LPP算法。

关键词: 人脸识别, 特征提取, Haar小波变换, 保局投影, 最近邻分类

Abstract: For the use of Locality Preserving Projections(LPP) for face recognition and the real flow is damaged by noise, this paper presents a new method, named HaarLPP method, to solve the noise. The method uses Haar wavelet transformation to reduce the effect of noise. Face recognition is completed by the LPP dimensionality reduction and the nearest neighbor criteria. Experimental results based on AT&T and Sheffiled face database show that the method is superior to the traditional LPP algorithm in term of the noise sensibility.

Key words: face recognition, feature extraction, Haar wavelet transformation, Locality Preserving Projections(LPP), nearest neighbor classification

中图分类号: 

• TP391.41