Establishing Hyper-Renormalization for Geometric Estimation from Images

虽然极大似然已经用于从图像数据中进行高精度几何估计，例如拟合直线和椭圆以及计算多幅图像上对应点之间的关系，但我们推导了一种更高精度的方法，称为“超重正化”，并将其应用于各种实际问题。同时，我们对最大似然解的修正进行了“超精确校正”的推广，证实了可以达到同样的精度。本文还介绍了一种新的扩展FNS方法，并将其应用于东日本大地震GPS陆地形变资料的分析。

While maximum likelihood has been used for high accuracy geometric estimation from image data, such as fitting lines and ellipses and computing the relationships between corresponding points over multiple images, we have derived a higher accuracy method, called “hyper-renormalization” and applied it to various practical problems. At the same time, we have extend the “hyper accurate correction” for correcting the maximum likelihood solution and confirmed that the same degree of accuracy can be reached. We have also introduced a new formulation of the extended FNS method and applied it to the analysis of the GPS land deformation data of the Great East Japan Earthquake.