结合HMRF模型的模糊ISODATA高分辨率遥感图像分割

模糊ISODATA (Fuzzy ISODATA, FISODATA) 算法不但继承了FCM算法的可拓展性以及ISODATA算法的自组织性，还能自动获取数据的类属数，因此在许多数据处理领域中有着广泛的应用。在应用于图像分割时，FISODATA算法定义的FCM目标函数未考虑邻域像素间的数据相关性，导致该算法的抗噪性能较差；此外，FISODATA算法中分裂-合并操作需人工选取阈值参数，而不适当的阈值往往使得该算法陷入局部极值, 因而得到错误的类属数并影响图像分割结果。该文将考虑邻域关系的基于隐马尔可夫随机场（Hidden Markov Random Field，HMRF）的FCM（HMRF-FCM）方法纳入ISODATA框架，提出HMRF-FCM ISODATA (HMRF-FISODATA)算法，在分裂与合并操作后增加了优化操作，并根据优化结果自适应调节控制聚类分裂与合并的各阈值。该算法不仅能够快速获取正确类属数，而且克服了FISODATA算法没有考虑邻域像素的关系、人工选取阈值参数和受图像噪声影响大等问题，实现了自动确定正确确定类属数的同时完成高精度图像分割。 

A Fuzzy ISODATA Approach Combing Hidden Markov Random Field Model for High Resolution Remote Sensing Image Segmentation

Fuzzy ISODATA (FISODATA) algorithm inherits the expansibility of FCM and self-organization of ISODATA and can obtain the number of classes and good clustering results simultaneously. Consequently, FISODATA has been already applied in many image processing fields. For image segmentation tasks, its objective function does not consider the effects from neighbor pixels, so it is sensitive for noises. Besides, the splitting and merging operations of FISODATA need parameters selected manually which may cause local optimization result. In this paper, Hidden Markov Random Field FCM (HMRF-FCM) is brought into the ISODATA framework, the adaptive splitting and merging operations are designed for the purpose of formulating HMRF-FCM ISODATA (HMRF-FISODATA) algorithm. The proposed algorithm can not only obtain the correct number of classes automatically but also overcomes the shortcoming of manually selecting parameters and seriously noise effect of FISODATA, as a result, it obtains segmentation results accurately.