SIFT算法优化及其在遥感影像配准中的应用

Improved SIFT Algorithm and Its Application in Registration of Remotely Sensed Imagery

针对传统尺度不变特征转换(scale invariant feature transform,SIFT)方法在处理存在角度偏差的图像配准数据时得到的配准点对数量低以及配准精度不高的问题,提出一种基于多角度归一化互相关法优化的SIFT遥感图像配准方法。以相关性系数为标准确定图像最佳配准位置,进行角度校正;用SIFT算法进行特征提取和特征匹配,并结合随机抽样一致性算法(random sample consensus,RANSAC),剔除错误配准点,以提高配准精度。实验表明,该实验配准方法比单一的SIFT配准方法得到数量更多且精度更高的特征点对,结果显示SIFT配准点对数量平均提高24.5倍,RANSAC算法确定的正确配准点对平均调高86.8倍。

An improved Scale Invariant Feature Transform（SIFT）automatic registration of high-spatial resolution remote sensing images based on multi-angle Normalized Cross Correlation（NCC）is proposed in this paper to solve theproblem that the traditional SIFT method has low registration point and low registration accuracy when dealing with the image registration data with angle deviation.Fristly,determine the best image registration position based on correlation coefficient,and complete the rotation offset correction.Then,SIFT is used to extract the feature points and feature matching,which is subsequently refined by Random Sample Consensus（RANSAC）to eliminate the mismatched control points.Experimental results indicated that this registration method gives higher precision matched points than simple SIFT and the number of points average increasing 24.5times,and the refined points by RANSAC average increasing 86.8times.