摘要
论文提出了基于小波特征的核主分量分析技术,即在进行非线性映射之前,首先利用小波变换对原始输入训练样本进行预处理,获取低频平滑、水平细节和垂直细节等三个子图的小波特征,然后在频域上,对它们分别进行核主分量分析(KPCA),对最终获得的3组特征向量设计了一种特征融合的方法。在ORL标准人脸库上的试验结果表明所提方法不仅在识别性能上优于现有的核主分量分析方法,而且,特征抽取速度提高了11倍。
A novel kernel principal component analysis method based on wavelet feature is developed in the paper.Wavelet transform is first employed to preprocess the original training samples and threee groups of wavelet features,which correspond to lower frequency,horizontal detail and vertical detail respectively,are obtained.Kernel principal com-ponent analysis(KPCA)is then performed in each transformed lower dimensional samples.In order to combine three classes of nonlinear principal component features obtained above together,a feature fusion method is presented.Finally,the experimental results on ORL face database indicate that the proposed method is superior to KPCA in the recogniton rate and11times faster than KPCA in feature extraction.
出处
《计算机工程与应用》
CSCD
北大核心
2004年第22期45-47,93,共4页
Computer Engineering and Applications
基金
国家自然科学基金(编号:60072034)
国家教委博士点基金资助
关键词
核主分量分析
小波分解
特征抽取
人脸识别
kernel principal component analysis,wavelet decomposition,feature extraction,face recognition
