The class statistics and Euclidean measure are both used to determine the neighborhood of a pixel to keep classes separability and preserve data distribution.

• 我们采用欧式距离及资料相似度两种度量，目的是为了兼顾资料原有之几何架构与类别分离度。