特征描述和特征匹配是计算机视觉领域的重要组成部分.近年来,为了实现图像匹配上的可靠性和鲁棒性,许多特征描述算法被提出来,例如SIFT、SURF、DAISY和BRIEF等.然而,当图像发生平移、旋转、缩放等大视角变化时,这些描述符通常会失效.为了解决这个问题,在局部近邻图模型的基础上,提出一种新颖的特征描述和相似性度量方法(LNFM算法).所提出的描述符和相似度可以很好地应用于各种流行的图像匹配算法.实验结果表明:在特征匹配过程中,该算法可以检测到可靠的匹配关系,性能较为优越.%Feature description and feature matching are important parts in the field of computer vision.In recent years, many feature description algorithms have been proposed to achieve reliable and robust performance in image matching, such as SIFT, SURF, DAISY and BRIEF.However, their descriptors usually fail when the images have undergone large viewpoint changes such as translation, rotation and scaling.To solve this problem, on the basis of local neighborhood graph model, a novel feature description and similarity measure method (referred to as LNFM algorithm) is proposed.The proposed descriptor and similarity can be well applied to a variety of popular image matching algorithms.The experimental results show that: in the process of feature matching, the proposed algorithm can detect reliable matching relationship, and the performance is relatively superior.
展开▼
