基于分水岭谱聚类的SAR图像分割.pdfVIP

27 6 Vo.l 27, No. 6 2008 12 J. Infrared M illmi . W aves Decem ber, 2008 : 1001- 014( 2008) 06- 0452- 05 - SAR 1 2 马秀丽 , 焦李成 ( 1. , 200072; 2. , 710071) : 由于谱聚类是基于图论的以相似性为基础的聚类方法, 需要计算图像中每 像素点之间的相似性. 当图像 很大时, 计算相似性矩阵和求解相应的特征值特征向量是很困难和耗时的. 为此, 针 合成孔径雷达 ( SAR) 图像 的特点, 提出了 一个两阶段的图像分割方法, 首先采用分水岭算法 图像进行过分割, 然后再用改进的谱聚类算法 进行聚类. 新方法不仅可以减少噪声 分割结果的影响, 很好地保持图像边缘, 而且 时间要求较高的应用也具有 一定的参考价值. 为了验证新方法的有效性, 将其用于 SAR 图像分割, 取得了较优的分割结果. : 谱聚类算法; 分水岭; SAR 图像分割 : TP75 : A SAR IMAGE SEGMENTATION BASED ON WATERSHED AND SPECTRAL CLUSTERING 1 2 MA X iu-L i, JIAO L-iCheng ( 1. School of Commun ication and Inform ation Engineering, ShanghaiUn iversity, Shangha i 200072, Ch ina; 2. Institute of Intelligent Inform ation Processing, X idian Un ivers ity, X i. an 710071, Ch ina) Abstract: Spectral clustering is a smi ilarity-based clustering algorithm on graph theory. W hen the processed mi age is very huge, it is very difficu lt and tmi e-consum ing to com pute affin ity m atrix and its eigenvalues and eigenvectors. A mi ing at the characteristics of synthetic aperture radar( SAR) mi ages, a two-stage mi age segmentation algorithm w as proposed, in wh ich w atershed was used to produce over-segmentation and an mi proved spectral clustering algorithm was app lied to perform final clustering. The new algorithm can not only reduce the noise in SAR mi ages and keep their boundary very w ell but also is valuable to