| 基于相似特征点集的SIFT匹配改进算法 |
| |
| 引用本文： | 高放,陆频频,王旭.基于相似特征点集的SIFT匹配改进算法[J].测绘工程,2016,25(6):19-23. |
| |
| 作者姓名： | 高放 陆频频 王旭 |
| |
| 作者单位： | 辽宁工程技术大学 测绘与地理科学学院,辽宁 阜新,123000;辽宁工程技术大学 测绘与地理科学学院,辽宁 阜新,123000;辽宁工程技术大学 测绘与地理科学学院,辽宁 阜新,123000 |
| |
| 摘 要： | 当影像中存在相似或重复场景时,传统SIFT匹配算法存在匹配成功率低,目前改进的SIFT匹配算法计算量大。基于相似特征点集的SIFT匹配改进算法,依据相似性或重复场景的影像纹理特点,在SIFT特征点匹配过程中,通过设定阈值提取初始同名点,建立针对未成功匹配参考特征点的相似特征点集,利用已获取初始同名点建立仿射几何约束模型构建参考特征点的匹配约束窗口,在该窗口内利用特征点相对主方向及尺度约束,对特征相似点集进行匹配获得同名点,最后采用RANSAC算法剔除误匹配点。对比实验结果表明,在影像像对间存在较多相似性场景,同时存在较大尺度缩放、旋转变换、视角及模糊差异的情况下,文中算法在匹配成功率和计算复杂度上具有明显的优势。
|
| 关 键 词： | SIFT 相似特征点集 几何约束 匹配 初始同名点 |
Improved sift matching algorithm based on similar feature point set |
| |
| GAO Fang,LU Pinpin,WANG Xu.Improved sift matching algorithm based on similar feature point set[J].Engineering of Surveying and Mapping,2016,25(6):19-23. |
| |
| Authors: | GAO Fang LU Pinpin WANG Xu |
| |
| Abstract: | Considering the low matching rate of traditional SIFT algorithm and large amount of calculation for the present improved SIFT matching method w hen similar or repeat scene existing in images ,a new improved SIFT matching method is proposed based on similar feature point set in this paper .Firstly ,initial matching points are extracted by threshold during traditional SIFT mateching method ,at the same time the similar feature point set is built for the reference point feature w hich can not be matched .Secondly , geometrical constraint model is constructed by the initial homonymy points ,w hich is used to build constraint window for reference point feature .Then the relative scale and main orientation of SIFT features are utilized to extract the homonymy points .At last ,RANSAC is imbedded to eliminate mismatching points .Compared with other matching algorithms ,the proposed algorithm has significant advantages in terms of successful matching rate and computational complexity . |
| |
| Keywords: | SIFT (scale inveriant feature transform ) similar feature point set geometrical constraint matching initial homonymy point |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
