Bonnet N, Liehn J C
Laboratoire de Microscopie Electronique (Unité INSERM 314) BP 347-Reims, France.
J Electron Microsc Tech. 1988 Sep;10(1):27-33. doi: 10.1002/jemt.1060100105.
The geometric registration of two electron microscopic images generally is performed by maximizing the cross-correlation coefficient between them. We show that a new similarity measure (the number of sign changes) is useful for performing simultaneously geometric and gray-level registration. This method is robust, which means that it provides a good estimation of the parameters even in the presence of outliers that cannot be described by the registration model.
两个电子显微镜图像的几何配准通常是通过最大化它们之间的互相关系数来实现的。我们表明，一种新的相似性度量（符号变化的数量）对于同时进行几何和灰度配准很有用。该方法具有鲁棒性，这意味着即使存在无法用配准模型描述的异常值，它也能对参数进行良好的估计。
