基于四元数矩阵奇异值分解的彩色图像识别

本文首先以实验证明了彩色图像矩阵的奇异值（SVS）仅含有图像的少量信息，大量信体现在图像矩阵的奇异值分解（SVDQ）的两个四元数酉矩阵中。然后给出了一种新的图像特征提取方法。该方法将图像投影到SVDQ的各个正交基上，得到投影系数向量。将此向量作为图像的代数特征并用于彩色图像识别中。实验表明，与奇异值特征向量用于彩色图像识别方法相比，本文方法显著提高了识别率。

The Recognition of Color Images Based on the Singular Value Decompositions of Quaternion Matrices

In this paper, with the experiment in color image databases, it is shown that the SVs contain little information of color image and most important information is contained in the two quaternion unitary matrices of SVDQ. So, a new feature extraction method for color image has been proposed. Firstly, the image is projected on to the orthogonal basis of SVDQ, then the projection coefficient vector is used as algebraic feature of image and applied to recognition. With the experiments for image recognition, it is shown that the recognition rate is increased compared to the method by the singular feature vector.