一种基于特征点的图像匹配算法  

An automatic image stitching algorithm based on feature matching

作　　者：谭磊[1] 张桦[1] 薛彦斌[1] 

机构地区：[1]天津理工大学计算机科学与技术学院,天津300191

出　　处：《天津理工大学学报》2006年第6期66-69,共4页Journal of Tianjin University of Technology

基　　金：天津市高等学校科技发展基金(2004BA09)

摘　　要：本文提出一种图像特征点匹配算法,并在该算法的基础上形成构建全景图的图像拼接算法.此算法采用Harris角检测算子进行特征点提取,并为其分配特征描述符.在进行相邻图片的特征比对时,提出一种基于小波系数的特征索引算法,提高搜索效率.运用稳健的RANSAC算法将伪匹配点集合划分成为内点与外点,在内点域中精确计算图像之间的变换关系.算法的重要特点为:基于小波系数的特征索引,可以使不同图像之间匹配特征点的搜索效率显著提高.实验结果表明:该算法得到的匹配点精确,受图像的形变、噪声影响较小;图像拼接处理的效果较好,具有较高的实用价值.This paper presents an image feature matching algorithm, based on what an image stitching algorithm can be functioning. In this algorithm, Harris comer detector is used to extract feature points. And each feature point is assigned a feature descriptor. A feature indexing algorithm based on wavelet coefficients is used when comparing features in neighboring images, which increases efficiency in the nearest neighbor searching. Then, a pseudo matching set is divided into inliers and outliers using robust RANSAC algorithm. The accurate transformation matrix between two images can be estimated within the inliers subset. The important characteristics is shown in this algorithm. The algorithm of feature indexing based on wavelet coefficients will speed up the search of matched point pairs. Experiments show that feature matches are accurate and robust towards distortion and noise. And high quality image stitching results can be produced. Thus the algorithm is highly valuable in practice.

关 键 词：特征点 小波变换 K-近邻 图像拼接 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]