摘要
提出了一种采用小波包重构系数矩阵与改进SVD的人脸识别新算法。小波包变换是小波变换的推广,可视为普通小波函数的线性组合,具有灵活的时频分析能力。小波包重构系数矩阵与原始图像矩阵的尺寸相同,具有较高的精度。使用常规Colub-Reish算法的奇异值分解(SVD)所得到的奇异值(SV)按由大到小的顺序重新排列过,无法确定每个SV与输入矩阵列向量的对应关系。改进的SVD方法能够使得奇异值与每个频带的重构系数相对应,进而构造出人脸图像小波包重构系数矩阵的奇异值特征向量,并采用基于方差计算的相似度分类方法识别人脸。实验表明,该方法识别率高、稳定性强。
A face recognition algorithm is proposed based on wavelet packet re-constructed coefficient matrixes and improved singular value decomposition（SVD）.Wavelet packet transform derived from wavelet transform can be considered as a linear combination of general wavelet functions and has flexible time-frequency analysis ability.Wavelet packet re-constructed coefficient matrix has the same size as the original image matrix,so it has high accuracy.Singular value（SV）,attained by traditional SVD based on Colub-Reish algorithm is rearranged in the light of big-small order,so that the corresponding relationship between each SV and column vector of imput matrix is unidentified.The improved SVD method is used to construct SV feature vectors of wavelet packet re-constructed coefficient matrixes for facial images,and it makes SV and re-constructed coefficient of each frequency band relative.Face recognition is realized by classification method using variance similarity degree.The experiments show that the proposed method has the characteristics of high recognition rate and good stability.
出处
《控制工程》
CSCD
北大核心
2010年第5期629-631,635,共4页
Control Engineering of China
基金
国家教育部博士点基金资助项目(2006021600)
关键词
人脸识别
小波包变换
改进SVD
SV特征向量
方差相似度
face recognition
wavelet packet transform
improved SVD
SV feature vector
variance similarity degree
