基于极大似然估计采样一致性准则的遥感影像配准参数解算方法研究

Resolving method of remote sensing image registration coefficient based on Maximum Likelihood Estimation Sample Consensus

本文提出运用最大似然采样一致性准则解算遥感影像配准系数的方法。该方法基于极大似然估计理论,首先对初始匹配点的坐标残差进行概率建模,计算概率模型成立时的似然函数值并选择似然函数值最大时的参数为正确结果,最终剔除错误点保留正确匹配点。该方法较之传统的最小二乘方法更为准确地计算配准系数,并可以解决随机采样一致性准则解算配准参数时,对阈值的依赖问题。试验证明,该方法可提高配准参数解算的稳健性和精度。

This paper proposed an resolving method of image registration coefficient which utilizes Maximum Likelihood Estimation Sample Consensus （MLESAC）. The method originates from Maximum Likelihood Estimation theory. At first, it made probablity modeling of initial correspondence points. And then it calculated the value of likelihood function assuming the model was correct. At last, MLESAC algorithm selected the coefficient which has the max value of likelihood function as final result, and removed wrong match points. Comparing with least squared method, MLESAC algorithm could be able to calculate registration coefficient more precisely. Furthermore MLESAC algorithm is independent of threshold, and more robust than random sample consensus （RANSAC）. Experiments in the paper proved that, MLESAC improved the resolving robustness and precision of image registration coefficient.