2DPCA-ICA算法在人脸识别中的应用

Applications of 2DPCA-ICA in face recognition

传统独立元分析(Independent Component Analysis,ICA)用于人脸识别首先是将人脸图像矩阵转换成向量求白化矩阵,然后利用快速固定点算法求分离矩阵,获得人脸图像独立基子空间,从而实现人脸识别。二维主元分析(Two-dimensional Principle Component Analysis,2DPCA)无须将人脸图像矩阵转换成向量,直接利用二维人脸图像矩阵求协方差矩阵,其特征值与特征向量的计算得到简化。本文结合2DPCA与ICA算法的特点,提出2DPCA-ICA人脸识别算法。该方法通过2DPCA算法计算白化矩阵;接着利用ICA算法获得人脸图像的独立元;然后构造独立基子空间;最后依据测试样本在独立基子空间上的投影特征实现人脸识别。基于ORL与Yale人脸数据库的实验结果表明,2DPCA-ICA算法正确识别率与识别效率均高于PCA-ICA算法与2DPCA算法,是一种有效的人脸识别方法。

In face recognition traditional Independent Component Analysis （ICA） is to convert face image matrix into vector to find whitened matrix, and separate matrix is solved by way of Fast ICA. Thus independent basis subspace of face image is obtained, and face recognition is realized. Two-dimensional Principle Component Analysis （2DPCA） is used to compute covariance matrix directly according to two-dimensional matrix of face image, which is not be transformed into vector, and computation of eigenvalues and eigenvectors are predigested. Combined with the characteristics of 2DPCA and ICA, a novel method for 2DPCA-ICA in face recognition is presented in this paper. As opposed to PCA-ICA, whitened matrix is firstly computed through 2DPCA, and independent components of face image are obtained. Then independent basis subspace is constructed. Finally, face recognition is finished using projection features of test sample on independent basis subspace. Experimental results on ORL and Yale face databases show that 2DPCA-ICA has the advantages over PCA-ICA and 2DPCA in correct recognition rate and recognition efficiency, and is valid in face recognition.