基于空间变换迭代的SIFT特征图像匹配算法  

SIFT matching algorithm with geometry constraint

作　　者：杨新锋[1] 滕书华[2] 夏东[2] 

机构地区：[1]南阳理工学院计算机与信息工程学院,河南南阳473004 [2]国防科学技术大学电子科学与工程学院,湖南长沙410073

出　　处：《红外与激光工程》2013年第12期3496-3501,共6页Infrared and Laser Engineering

基　　金：中国博士后科学基金(2012M512168);河南省科技攻关项目(122102210563;132102210215)

摘　　要：SIFT(Scale Invariant Feature Transform)特征由于具有旋转、平移和尺度不变性在图像匹配中得到了广泛的应用。但直接运用SIFT特征进行匹配,存在两个问题:易受匹配参数的影响,出现较多的错漏匹配现象;只适用于相似变换情况下的图像匹配,对于高维的仿射变换情况则难以奏效,而在实际图像匹配中这种情况更为常见。针对以上问题,提出了一种空间变换迭代的SIFT特征图像匹配方法。把SIFT特征点集匹配转化为SIFT特征向量与点集的几何分布信息相关的函数最优化求解问题,通过在确定性退火框架下,迭代求解空间仿射变换与点集匹配对应关系,最终得到最优的SIFT特征点匹配关系。仿真实验表明:在较大仿射变换情况下该方法仍能实现图像SIFT特征点集的正确匹配。For the rotation, translation, scale invariant properties of SIFT （Scale Invariant Feature Transform） feature, it has been widely applied in imaging matching. But there are two defects of using SIFT while matching. Firstly, the matching performance is directly affected by the matching parameters, and there is always mismatching and error matching existed. Secondly, it only fits for matching under similarity transformation, while at the affine transformation situation it fails. In this paper, a novel iterative matching algorithm based on transformation estimation was proposed. The SIFT matching problem was turned into an optimization problem about SIFT feature vector and the geometry distribution of the point sets. By searching for the affine transformation and correspondences under the iterative deterministic annealing frame, the algorithm got the optimal matching result of SIFT point sets. Experiment results show that even at large affine transformation, the algorithm can still get the right matching results.

关 键 词：局部不变特征 仿射变换 SIFT 图像匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]