基于马氏距离谱特征的图像匹配算法

Image Matching Algorithm Based on Mahalanobis-Distance Spectral Features

传统的基于谱特征的图像匹配算法中,采用的欧式距离度量不能公平地反映数据样本各维度分量之间的潜在关系,并且当存在较大的形变和出格点时匹配精度和稳定性较差.为了解决谱特征构造中所存在的问题,文中提出一种基于马氏距离谱特征的图像匹配算法.该算法首先利用马氏距离在子特征点集上构造局部无向加权图;接着对图的关联邻接矩阵进行奇异值分解,用特征值向量构造描述点集属性的马氏距离谱特征;然后根据马氏距离谱特征构造出匹配矩阵,并利用贪心算法得到图像特征点之间的匹配关系;最后,为了进一步提高匹配的精度,采用SVM方法剔除误匹配点.大量实验结果表明,该算法提高了匹配的精度,并且对出格点问题具有较高的鲁棒性.

In the traditional image matching algorithms based on spectral features,the Euclidean distance metric can not fairly reflect the underlying relationship between the dimensions of sample data,and large deformation and outliers will lead to poor matching accuracy and stability. In order to solve these problems existing in the structure of spectral features,an image matching algorithm based on Mahalanobis-distance spectral features is proposed. In the algorithm,first,a local undirected weighted graph is constructed in sub feature point sets by using the Mahalanobis distance. Next,the singular value decomposition of the adjacent matrixes of the graph is performed,and the Mahalanobis-distance spectral features describing the attributes of the point sets are constructed by using spectral value vectors. Then,a matching matrix is constructed based on the Mahalanobis-distance spectral features,and the matching relationships among the feature points of the image are obtained by using the greedy algorithm.Finally,in order to further improve the matching accuracy,the false matching points are eliminated by means of the SVM method. A large number of experimental results show that the proposed algorithm improves the matching accuracy,and it is robust to outliers.