论文部分内容阅读
目的传统FCM算法及其改进算法均只采用隶属度作为分割判据实现图像分割。然而,在分割过程中聚类中心易受到同质区域内几何噪声的影响,导致此类算法难以有效分割具有几何噪声的图像。为了解决这一类问题,提出一种利用包含度和隶属度的遥感影像模糊分割算法。方法该算法假设同一聚类对每个像素都有不同程度的包含度,将包含度作为一种新测度来描述聚类与像素间关系,并将包含度纳入目标函数中。该算法通过迭代最小化目标函数来得到最优的隶属度和包含度,然后,通过反模糊化隶属度和包含度之积实现带有几何噪声的遥感图像的分割。结果采用本文算法分别对模拟图像,真实遥感影像进行分割实验,并与FCM算法和FLICM算法进行对比,定性结果表明,对含有几何噪声的区域,提出算法的用户精度和产品精度均高于FCM算法和FLICM算法,且总精度和Kappa值也高于对比算法。实验结果表明,本文算法能够抵抗几何噪声对图像分割的影响,且分割精度远远高于其他两种算法的分割精度。结论提出算法通过考虑聚类对像素的包含性,能够有效抵抗几何噪声对图像分割的影响,使得算法具有较高的抗几何噪声能力,进而提高该算法对含有几何噪声图像的分割精度。提出算法适用于包含几何噪声的高分辨率遥感图像,具有很好的抗几何噪声性。
Purpose The traditional FCM algorithm and its improved algorithm only use membership as the segmentation criterion to achieve image segmentation. However, the clustering centers are easily affected by the geometric noise in the homogeneous region during segmentation, which makes it difficult for such algorithms to effectively segment the image with geometric noise. In order to solve this kind of problem, a remote sensing image fuzzy segmentation algorithm using inclusion degree and membership degree is proposed. Methods The algorithm assumes that the same cluster has different degrees of inclusion for each pixel. The inclusion degree is used as a new measure to describe the relationship between clusters and pixels, and the inclusion degree is included in the objective function. The algorithm obtains the optimal degree of membership and inclusion by iterative minimization of the objective function. Then, the segmentation of the remote sensing image with geometric noise is achieved by the product of anti-fuzzy membership and inclusion degree. Results Compared with FCM algorithm and FLICM algorithm, the proposed algorithm is used to segment the simulated images and the real remote sensing images respectively. The qualitative results show that the user accuracy and product accuracy of the proposed algorithm are higher than those of the FCM algorithm And FLICM algorithm, and the total accuracy and Kappa value is also higher than the contrast algorithm. The experimental results show that the proposed algorithm can resist the influence of geometric noise on image segmentation, and the segmentation accuracy is much higher than the other two algorithms. The conclusion is put forward that the algorithm can effectively resist the influence of geometric noise on image segmentation by considering the inclusion of pixels in the clustering algorithm, so that the algorithm has high anti-geometric noise ability and improves the segmentation accuracy of the algorithm with geometric noises. The proposed algorithm is suitable for high resolution remote sensing images containing geometric noises and has good anti-geometric noise performance.