基于MRF模型和EM算法的多源图像融合方法

提出了两种图像融合方法.该方法首先利用EM-MRF算法与模糊分类方法的等价性,将EM-MRF算法引入到图像融合领域.在此基础上,利用统计模型对图像进行非监督分类的模型参数估计转化通过EM算法从不完全数据中估计模型参数的问题,并利用Markov随机场模型建立类别的先验概率、EM迭代算法进行图像分类的方法有较高的分类精度和鲁棒性,导出了基于分布式和集中式多传感器图像融合模型的两种融合方法.最后仿真试验表明,这两种融合方法既可以提高分类精度,又可以加强对噪声的抗干扰能力.

Multisource Image Fusion Method Using Markov Random Field Model and EM Algorithm

Two methods for feature fusion of remotely sensed image are presented. Expectation Maximization (EM)__Markov Random Field (MRF) algorithm is introduced to image fusion, taking advantage of the equivalence relation between EM-MRF and fuzzy classification algorithms. Distributed and centric image fusion methods are deduced respectively by using model parameter estimation in an unsupervised statistical model-based approach to transform the problems of parameter estimation from incomplete data and MRF model-based EM algorithm to improve classification accuracy and robustness. The realization and simulation experiment results show that the proposed methods can improve classification accuracy and enhance the ability of resisting noise interference.