结合HMRF模型的模糊ISODATA高分辨率遥感图像分割

发布时间：2018-03-08 11:18

本文选题：非监督分类　切入点：隐马尔可夫随机场　出处：《信号处理》2016年02期 　论文类型：期刊论文

【摘要】：模糊ISODATA(Fuzzy ISODATA,FISODATA)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)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,实现了自动正确确定类属数的同时完成高精度图像分割。
[Abstract]:Fuzzy ISODATA(Fuzzy ISODATAA (FISODATA) is applied to image segmentation. The FCM objective function defined by FISODATA algorithm does not take into account the data correlation between neighboring pixels, which results in poor anti-noise performance of the algorithm. In addition, the splitting and merging operation in FISODATA algorithm requires manual selection of threshold parameters. However, improper threshold often makes the algorithm fall into local extremum, so it gets wrong generic number and affects image segmentation results. In this paper, Hidden Markov Random FieldFCM-HMRF-FCM-HMRF-FCM-based Hidden Markov Random FieldFCM-HMRF-FCM-based method, which takes into account neighborhood relationship, is incorporated into the framework of ISODATA. The HMRF-FCM ISODATAA HMRF-FISODATAA algorithm is proposed, which increases the optimization operation after splitting and merging operations, and adaptively adjusts the threshold of clustering splitting and merging according to the optimization results. The algorithm overcomes the problem that FISODATA algorithm does not consider the relation of neighborhood pixels, manually selects threshold parameters and is greatly affected by image noise, and realizes automatic and correct classification and high precision image segmentation.
【作者单位】：辽宁工程技术大学测绘与地理科学学院遥感科学与应用研究院;
【基金】：国家自然科学基金青年科学基金项目(41301479);国家自然科学基金面上项目(41271435)
【分类号】：TP751

【参考文献】