改进尺度不变特征变换算法的图像配准

Improved SIFT Algorithm for Image Registration

为了提高尺度不变特征变换(SIFT)算法的配准精度和适应性,对匹配框架和匹配方法进行改进。匹配算法优化为初始匹配、中间匹配及精确匹配三个阶段,首先利用最近邻值、最次近邻比值对SIFT特征点进行分离得到两类初始匹配点集;中间匹配阶段对Ⅰ类匹配点集进行初步提纯并将一对多、多对一及双向匹配不一致点对归入Ⅱ类匹配点集;为提高匹配效率改进对极几何约束匹配方法来剔除误匹配点对完成精确匹配。结果表明,提出的改进算法在抗几何变形和配准精度等方面具有优越性。

In order to improve the accuracy and adaptability of SIFT algorithm,the matching framework and matching method are improved.The matching algorithm is optimized for initial matching,middle matching and precision matching.Firstly the SIFT feature points are separated by the nearest neighbor value and the nearest neighbor ratio value,and the dual initial matching points are gotten.In the process of middle matching,the middle matching points are made up of a preliminary purification matching points and the second matching points including one-to-many,many-to-one and inconsistent bi-direction matching points.To improve the matching efficiency,an improved epipolar geometry constraint method is used to eliminate false matching points in the process of precision matching.The results show that the proposed improved algorithm has better performance in terms of registration accuracy and robustness to geometric deformation within images.