基于改进JSEG算法的高分辨率遥感图像分割方法  

High Resolution Color Remote Sensing Image Segmentation Based on Improved JSEG Algorithm

作　　者：冯晓毅[1] 王西博[1] 王蕾 彭进业[1] 

机构地区：[1]西北工业大学电子信息学院 [2]63620部队

出　　处：《计算机科学》2012年第8期284-287,共4页Computer Science

基　　金：国家自然科学基金(NSFC-60875016);新世纪优秀人才支持计划(NCET-08-0458)资助

摘　　要：JSEG算法是一种有效的彩色图像分割方法,但该方法直接用于遥感图像分割时,往往会出现因遥感图像区域边界较模糊而导致区域边界分割不准确,或因区域阴影而导致的过分割现象。为了解决上述问题,提出基于改进JSEG算法的遥感图像分割方法,该方法利用能更好描述区域内颜色的同质性的局部同质矩阵来校正传统JSEG算法中的局部J值,以实现对区域边界的准确反映,提高区域边界分割的准确性;利用图像的LBP/C纹理特征,合并具有相似纹理信息的颜色类,以减弱传统JSEG算法的过分割现象。仿真实验验证了上述方法的有效性。JSEG algorithm is a method of segmentation of color images which can make the image filtering,color space quantization and spatial segmentation in due order.However,using the method directly in segmentation could result in inaccurate of edge due to fuzzy edge in the remote sensing image,or lead to excessive segmentation because of different shades in the region.To obtain the better effect of segmentation,the local homogeneous matrix which gives reasonably description for color homogeneity in region and the boundary of different region was used to correct the local J value of traditional JSEG algorithm in this paper.In addition,to weaken or to eliminate the over segmentation,LBP/C operator which gives stable description for the texture information of image was used to combine class maps with similar texture information.Experimental results show that the above methods work well.

关 键 词：高分辨率遥感图像 JSEG算法 局部同质 LBP/C算子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]