应用于高分辨率遥感影像配准的SIFT特征均匀分布算法

A Uniformly Distributed SIFT Algorithm for Registration of High-resolution Remote Sensing Images

高分辨率遥感影像配准要求提取出稳定且分布均匀的同名点，但是传统的SIFT算法没有考虑特征点的空间分布情况，提取的特征点往往分布不均匀。针对高分辨率遥感影像数据的特点，提出了一种自适应控制SIFT特征均匀分布算法，利用局部纹理特征，并结合最优化筛选策略，在保证特征点稳定性和准确性的同时，控制特征点在不同空间的分布情况，实现sIFT特征点在图像空间和尺度空间的合理分布。分别采用了常规和多视角航空高分辨率遥感影像进行试验，试验结果表明，算法能够有效控制提取特征点的分布，具有一定的抗噪声干扰能力。

Registration of high-resolution remote sensing images requires stable and uniformly distributed homologous points,but feature points extracted by traditional scale-invariant feature transform（SIFT） algorithm are usually unevenly distribu- ted for lack of consideration of spatial distribution. Based on local texture features, an auto-adaptive uniformly distributed SIFT matching algorithm for high-resolution remote sensing images is proposed, and an optimization selection strategy is adopted to guarantee the reliability and accuracy and control the spatial distribution of feature points. As a result, the SIFT feature points can be uniformly distributed both in image space and scale space. Experiments are conducted with conventional and multi-view high- resolution remote sensing images respectively, and result shows that the algorithm can effectively control the distribution of feature points and resist noise interference.