基于局部不变特征点的高分辨率遥感影像配准

High-resolution Remotely Sensed Imagery Registration Based on Local Invariant Feature Points

为了解决传统的点特征匹配算法存在配准精度低的问题,本文提出一种基于局部不变特征点的高分辨率遥感影像配准算法。首先,利用双边滤波器对基准影像和待配准影像分别进行预处理,保留其边缘信息并去除噪声;其次,对预处理后的两幅影像进行局部不变特征点提取;然后,通过最近邻和次近邻的欧式距离比值法进行双向匹配,得到匹配特征点;最后,对待配准影像进行仿射变换。为了验证方法的有效性,提出一种仿射变换均方根误差AT-RMSE对精度进行评价,选取深圳市两个时相的PLEIADES影像,结果表明双边滤波器虽然增加了算法的运行时间,但能够降低55%以上的AT-RMSE;另外,基于SURF特征的配准方法比基于SIFT的性能更优,精度有大幅度提升。

To solve the problems of low registration accuracy of the traditional point feature matching algorithm, this article proposes a high - resolution remotely sensed imagery registration algorithm based on local invariant feature points. Firstly, the reference image and the to - be - registered one are preprocesscd respectively with the bilateral filter to retain the edge information and remove noise. Sec- ondly, local invariant feature points arc extracted from the two images. Then, matched feature points arc obtained through a bilateral matching by the ratio of Euclidean distances of the nearest neighbor to that of the next nearest one. At last, an affine transformation was carried out to the to - be - registered image. For verifying the validity of the proposed method, an accuracy assessment method was presented based on the Affine Transformation Root Mean Squared Error （AT - RMSE） , and experimental data is chosen from two phase PLEIADES images covering Shenzhen, China. The results show that the bilateral filter can reduce more than 55% of AT - RMSE though it raises the executing time of the proposed method. In addition, the performance of registration method based on SURF is better than that based on SIFT, and the accuracy is greatly improved.