一种改进的马尔可夫随机场遥感图像分割方法

Remote Sensing Image Segmentation Method Based on Improved Markov Random Field

对于传统马尔可夫随机场而言,先验能量的势能函数中的先验参数通常是根据经验手动选取大于零的值,没有考虑像元之间的距离,也没有充分考虑图像局部邻域先验特征,针对上述问题,提出一种结合标号场先验特征和像元距离动态估计先验参数的方法,并在先验能量中定义了观测场像元之间的影响系数,似然能量函数中引入Sobel算子描述观测场像元之间的关系,最后结合分水岭算法消除碎屑小区域进一步优化分割结果。通过Merced Land Use Dataset场景分类数据集进行了相关实验,结果表明该方法可以有效应用于遥感图像分割工作中。

For traditional Markov random field,the prior parameters in the potential energy function of prior energy are usually manually select-ed based on experience.The distance between pixels is not considered,and the local neighborhood prior features of image are not fully consid-ered.In view of the above problems,we proposed a method of dynamically estimating the prior parameters by combining the prior features of la-bel field and the distance of pixels,which defined the influence coefficient between the pixels of observation field in the prior energy,and intro-duced the Sobel operator into the likelihood energy function to describe the relationship between the pixels of observation field,and combined with the watershed algorithm to eliminate small areas of debris to further optimize the segmentation results.We carried out the relevant experi-ments on the scene classification dataset of Merced Land Use Dataset.The result shows that the method can be effectively applied to remote sens-ing image segmentation.