一种快速鲁棒的SAR图像匹配算法

Fast and Robust SAR Image Matching Algorithm

针对目前SIFT算法及其改进算法在多波段SAR图像匹配时匹配性能低下(普适性差、匹配精度低、时间复杂度高)的问题,在SIFT算法框架下分别从尺度空间构建和描述符构建两个方面进行改进。在构建尺度空间层面,提出将高斯引导滤波引入多尺度空间构建和预处理阶段,采用双边滤波策略,充分利用高斯引导滤波的实时性和旋转对称性与双边滤波的边缘保持优势,高效地滤除斑点噪声并保持边缘信息。在构建描述符阶段,提出采用局部差分二进制(Local Difference Binary,LDB)算法描述特征,在保证不降低特征点描述符区分性的同时,减少特征的向量维度,从而缩短构建描述符的时间。在特征匹配阶段,首先采用最近邻算法进行粗匹配,然后采用稀疏向量场一致性(Vector Field Consensus,VFC)快速剔除错误匹配点。实验结果表明,所提算法在SAR图像配准时间复杂度和匹配概率评价上要优于原始BFSIFT算法和KAZE算法。总体上,文中提出的SAR图像匹配算法是具有实时性、鲁棒性与高匹配概率的高效算法。

To solve the problem that SIFT and its improved algorithm have low matching performance（poor university,low matching accuracy,high time complexity）in the multi-band SAR image matching,we improved the algorithm respectively from creating scale space and descriptors within the framework of the SIFT algorithm.In scale space level,we proposed to use gauss guided filter to construct scale space and use bilateral filter in image pre-processing stage.This strategy,efficient filter speckles noise and keeps the image’s information,makes full use of gauss guided filter real-time and rotational symmetry and the edge preserving advantages of bilateral filter.In the construction descriptor stage,in order to ensure the distinction and reduce the time of build descriptors,we adopted the local difference binary to describing the local features’ characteristics.In the matching stage,the coarse matching uses the algorithm of nearest neighbor firstly,and then the sparse vector field consensus is used to remove the error matching points quickly.The experimental results show that the proposed algorithm from SAR image matching on time complexity and the matching probability is better than the BFSIFT and KAZE algorithm.In conclusion,our proposed algorithm is an efficient algorithm of real-time,robustness and high matching probability.