一种基于改进SIFT的航拍图像自动配准算法

摘要

针对传统SIFT算法匹配时间长、错匹配较多等问题,提出了一种基于改进SIFT特征的航拍图像自动配准算法.首先,通过特征点检测时设定检测极值点数目,按照DOG空间层次结构由粗到精来搜索特征点,并使用改进的SIFT特征描述符生成算法；其次,利用最近邻匹配准则进行初步匹配得到初始匹配点对,并采用双向匹配方法对匹配特征点对进行筛选；然后,基于马氏距离的特征点相似度量方法进行二次匹配,并使用RANSAC算法求取仿射变换模型；最后,通过双线性插值对变换后的图像进行重采样和插值.实验结果表明:该算法可以实现航拍图像之间的有效配准,在配准性能上优于传统SIFT算法.%To solve the problems of long registration time and many wrong registrations in using the traditional scale invariance feature transformation (SIFT) algorithm, an automatic aerial image registration algorithm is proposed based on the SIFT with improved features. Firstly, the feature point was searched according to a coarse-to-fine difference of Gauss (DOG) structure by setting a feature point number threshold, and the improved feature descriptors are used in the algorithm. Then the initial matching feature point pairs were obtained by use of the nearest neighborhood similarity measurement rule, and wrong registrations were removed by the lateral matching method. Mahalanobis distance was used to select right registrations in the second registration, and affine transformation was computed by Random Sample Consensus (RANSAC) algorithm. Finally, the transformed image was resampled and interpolated by means of bilinear interpolation. The experiments results show that the algorithm can achieve more accurate aerial image registration and is better than the traditional SIFT algorithm in performance.