基于动态拓展的特征匹配方法

Feature matching method based on dynamic expansion

针对图像间结构复杂、纹理重复的对象中存在特征匹配鲁棒性差、误匹配率高的问题,结合最近邻思想,提出一种基于动态拓展的特征匹配方法。输入待匹配的两幅图像,采用SIFT算子提取图像初始特征点几何位置信息,构建基础数据集;依据基础数据集,采用核心点周围邻域逐层约束的动态拓展聚类方法,划分图像聚类簇;设计度量函数确定图像对应聚类簇,生成对应簇内特征点的描述子;采用最近邻距离比准则进行特征匹配。以Oxford VGG标准数据集和古建筑图像为对象,验证了该算法的精确性与鲁棒性。

To solve the problems of the poor feature matching robustness and high false matching rate with complex structures and repeated textures,a feature matching method based on dynamic expansion was proposed combining with the nearest neighbors.For two images,the SIFT operator was used to extract the geometric position information of the initial feature points in images,and two basic data sets were obtained.Based on the core points in the basic data sets,a dynamic expansion clustering method was used combining with the layer-by-layer constraint,and the clusters in each image were obtained.The corresponding clusters between the left image and the right image were determined by the new measurement function,and the descriptors of the feature points in the corresponding clusters were generated.The nearest neighbors distance ratio criterion was used for feature matching.Taking the Oxford VGG standard data set and the ancient architectural images as objects,the accuracy and robustness of this algorithm are verified.