应用Hessian-Affine的遥感图像配准实验

Remote sensing image registration based on Hessian-Affine

针对存在云雾遮挡、仿射变形的遥感影像,本文提出应用Hessian-Affine与最大信息熵,检测并筛选仿射不变特征,同时选刺同名点估计初始变换参数,对每一个待匹配点预测出一定圆域约束内的对应匹配点集,利用NCC相关系数迭代确定圆域内真同名点,得到初始匹配点集,然后利用均方根误差(RMSE)迭代剔除误匹配,直至完成最佳仿射变换参数估计。结果表明:该算法对发生仿射畸变、气候复杂区域的影像配准表现出较好稳健性和定位精度。

Registration of multi-source remote ensing images is a tough problem in both image processing and computer vision, as it must deal with the block of clouds, affine transformation, etc. Accordingly, high-quality invariant registration primitive based on Hessian-Affine was put forward in this paper. The method could detect invariant features based on Hessian-Affine and filter invariant features based on information content of maximum. Simultaneously primary transformation parameters were estimated according to the tie points which were detected by handwork. The corresponding points within circular region restriction were predicted for every matc- hing point, and true matching point was determined by relevant coefficient iteration, so the initial set of matching points would be ob- tained. Then the root mean square error （RMSE） iteration was utilized to delete mismatching. Furthermore, affine transformation pa- rameters were calculated in this process. Eventually, image registration was achieved smoothly. Experiment results confirmed that the proposed algorithm has the robustness and good accuracy of registration for the images of affine deformation and poor weather condi-