摘要
针对单训练样本情况下人脸识别性能不佳的问题,本文提出了一种改进的基于奇异值扰动的人脸识别方法。首先通过奇异值扰动方法扩展人脸样本,然后运用小波变换压缩扩展样本,选择小波变换分解后的低频分量作为子图像,再采用核主成分分析提取人脸的高阶特征,最后根据最近邻分类器分类。在ORL和Yale数据库上的仿真实验证明了本文方法的识别性能优于对比方法。
In view of the poor performance of face recognition, an improved face recognition method based on singular-value-perturbed is proposed in this paper. Firstly, the singular-value-perturbed is applied to the single image so as to obtain expanded image set. Secondly, the wavelet decomposition is used as the pre-processing method, the low-frequency face image is chosen as a sub-image,and the high- order features are extracted by kernel principal component analysis. Finally, the nearest neighbor classifier is used for identification. The experiment results on ORL and Yale face databases show that the proposed method improves the recognition performance in comparison with the comparative approach.
出处
《计算机工程与科学》
CSCD
北大核心
2012年第10期88-91,共4页
Computer Engineering & Science
基金
湖北省自然科学基金资助项目(2009CDB069
2011CDC017)
湖北省教育厅创新团队资助项目(T201214)
关键词
人脸识别
特征提取
奇异值扰动
核主成分分析
小波变换
最近邻分类器
face recognition
feature extract
singular-value-perturbed
kernel principal component a-nalysis
wavelet transform
nearest neighbor classifier
