基于小波变换和非负矩阵分解的人脸识别改进方法  

An Improved Face Recognition Method Based on Wavelet Transform and Nonnegative Matrix Factorization

作　　者：杨绍华[1] 牛万红[2] 潘晨[3] 

机构地区：[1]宁夏大学数学计算机学院,宁夏银川750021 [2]宁夏大学远程教育学院,宁夏银川750021 [3]中国计量学院信息工程学院,浙江杭州310018

出　　处：《宁夏大学学报（自然科学版）》2010年第2期130-134,共5页Journal of Ningxia University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60663003);宁夏自然科学基金资助项目(NZ0823);宁夏大学科研基金资助项目((E)NDZR09-29;(E)NDZR09-47);宁夏大学"211工程"学科建设项目

摘　　要：针对NMF在识别人脸图像特征时分解速度慢,基空间不适应欧氏距离度量的缺点,提出了一种基于小波变换和非负矩阵基矩阵正交化的人脸识别方法.利用小波变换对人脸图像进行变换,选择LL分量既能抽取到人脸的实质特征又能有效减小数据维数,降低NMF分解的复杂性.同时对NMF的基矩阵实施正交化变换,在得到的正交基上进行投影.实验结果表明,该方法对光照变化、表情变化和部分遮罩不敏感,识别性能明显提高.Aiming at the lower resolving speed of NMF for face recognition and the disadvantages of space of bases not suitable for Euclid distance measuring,the method of face recognition with the use of wavelet transformation and the orthonormalized bases of nonnegative matrix is put forward.Making use of wavelet transformation to decompose the face images and choosing the LL coefficients,not only central facial features are captured but also computational complexity of NMF is reduced.Also,orthogonal transformation is put into base-matrix of NMF,and training samples and testing samples are projected onto orthogonal base-matrix.The experimental results show that the above-mentioned algorithm performs faster training speed and higher accuracy of recognition against the lower sensitivity to illumination,facial expression and part covering.

关 键 词：人脸识别 小波变换 非负矩阵分解 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]