基于局部空间信息的可变类模糊阈值光学遥感图像分割

阈值法分割在光学遥感图像分析中被得到广泛的应用,然而传统阈值法也存在诸多局限性,如对噪声敏感,需人为设定类别数,计算复杂度高等.针对传统闽值法的局限性,提出一种基于局部空间信息的可变类模糊阈值光学遥感图像分割方法.首先,以图像光谱的一阶矩为初始类中心,利用二分法原理和区域间最大相似度准则来快速确定类别数及其中心.然后,...

Fuzzy Threshold Optical Remote Sensing Image Segmentation With Variable Class Number Based on Local Spatial Information

Threshold segmentation has been widely used in optical remote sensing image analysis.However,traditional threshold methods also have many limitations,such as sensitivity to noise,artificially setting the number of classes,high computational complexity and so on.Aiming at the limitation of traditional threshold methods,a fuzzy threshold optical remote sensing image segmentation method with variable class numbers based on local spatial information is proposed.Firstly,taking the one-order moment of the image spectrum as the initial class center,the dichotomy principle and the maximum similarity criterion between regions are used to quickly determine the number of classes and their centers.Then,through the ridge-shaped fuzzy membership function,the degree of membership of each pixel to different classes is calculated.Meanwhile,considering the local spatial information of the membership of each pixel,a fuzzy weighted filter is defined in the membership domain to filter the membership matrix of each class.Based on the filtered membership set,the label field of the image is determined according to the maximum membership principle.Finally,the local abnormal labels in the label field are replaced,and the corrected label field is colored by the corresponding class center to obtain the segmented image.The results of visual and statistical analysis show that compared with the traditional threshold method,the proposed method can obtain better segmentation results while reducing the computation time.It can be applied to multi-threshold segmentation of optical remote sensing images.