一种改进的基于RANSAC方法的SIFT特征匹配

摘要

在图像配准,图像融合,目标识别,图像拼接等领域中,运用SIFT描述子进行特征匹配是一种常用方法,针对传统的消除误匹配的RANSAC算法迭代次数多,计算复杂度较大,在SIFT特征匹配中误配率高的问题,增加了一个原始匹配点对的前道处理环节,引入匹配点对的几何特征定义,根据匹配点对的最近邻点和次近邻点角度差在一定阈值范围内建立预处理模型,剔除误匹配点,减少了参与RANSAC的数据量,提高了RANSAC的运行效率.实验结果表明:与未增加前道处理的RANSAC算法相比,基本消除了误匹配点,提高了图像匹配的准确率.%In the field of image registration,image fusion,target recognition,image stitching,the use of SIFT descriptors for feature matching is a common method.Focusing on the problem of numerous iterations and complex computation,high error rate of RANSAC method in SIFT feature matching,this paper proposes to add a front-processing part to the original matching point pairs.The definition of geometric features of the matching pairs is introduced in this paper.A preprocessing model is established based on the nearest neighbor point and the next nearest neighbor point,because the angle difference is in a certain range.This can eliminate most of the mismatching points and reduce the amount of data involved in RANSAC.The experiments show that mismatching points are almost eliminated compared with the traditional RANSAC algorithm,thereby the algorithm improves the accuracy of image matching.