利用高斯混合模型的多光谱图像模糊聚类分割

摘要

针对传统分割算法难以实现高分辨率多光谱图像分割的问题,本文提出一种利用高斯混合模型的多光谱图像模糊聚类分割算法.该算法采用高斯混合模型定义像素对类属的非相似性测度,由于该算法具有高精度拟合数据统计分布能力,故可以有效剔除噪声对分割结果的影响.同时,引入隐马尔科夫随机场(Hidden Markov Random Field,HMRF)定义邻域作用的先验概率,并将其作为各高斯分量权值以及KL (Kullback-Leibler)信息中控制聚类尺度的参数,从而增强了算法对复杂场景遥感图像的鲁棒性,进一步提高了算法的分割精度.对模拟图像和高分辨多光谱图像分割结果进行了定性定量分析.实验结果表明:模拟图像的总精度达96.8％以上.这验证了本文算法在分割高分辨率多光谱图像时具有保留细节信息的能力,而且也证实了算法的有效性和可行性.该算法能够实现高分辨率多光谱图像的精确分割.%As the traditional segmentation algorithms are difficult to realize accurate segmentation for high resolution multispectral image,this paper proposed a kind of fuzzy clustering segmentation algorithm of multispectral image on the basis of Gaussian mixture model(GMM).GMM was adopted to define the dissimilarity measure of pixels to the clusters.As the proposed algorithm has the ability of high-precision fitting data statistics distribution,it can effectively eliminate the negative impact of noise on segmentation results.Hidden Markov Random Field (HMRF) was brought in to define prior probability of neighborhood relationship simultaneously,then the prior probability was used as weight of each Gaussian component and parameter to control cluster scale in Kullback-Leibler (KL),so the robustness of algorithm to remote sensing image at complex scene was increased and segmentation accuracy of algorithm was further improved.Qualitative and quantitative analysis were conducted on segmentation result of simulated image and high resolution multispectral image.Experimental result shows that the total accuracy of simulated image exceeds 96.8％,which verifies that algorithm mentioned has detailed information keeping capacity when performing segmentation to high resolution multispectral image and verifies effectiveness and feasibility of algorithm.The algorithm can realize accurate segmentation of the high resolution multispectral image.