【摘要】 将彩色人脸图像用四元数实矩阵的形式表示,对该矩阵作多级非抽样Contourlet(Nonsubsampled Contourlet,NSCT)变换,在各子带系数上作奇异值分解,进而获得图像的代数特征向量,由神经网络训练和识别。实验结果表明,该方法可以有效地提高彩色人脸图像的识别率,是一种有效的彩色人脸识别方法。更多还原

【Abstract】 In this paper,color face image is represented in quaternion real matrices.The coefficients in different scales and directions are obtained by image decomposition using the nonsubsampled contourlet transform.Singular value vector compressed is extracted as the color face image feature,and then the back propagation neural network is used to training and recognition.The experiment indicates that the recognition rate of the proposed method in this paper is improved and it is an effective color face image recognition method.更多还原