邻域级和超像素级差异图融合方法

Neighborhood-level and superpixel-level difference image fusion methods

考虑到传统代数运算法生成的合成孔径雷达(Synthetic Aperture Radar, SAR)图像的差异图变化区域和未变化区域对比度差,噪声鲁棒性弱,提出一种邻域级和超像素级差异图融合方法。首先通过对两期图像进行超像素分割,利用超像素结合对数运算获取超像素差异图;其次将相似性差异图和超像素差异图在邻域内进行融合;最后利用迭代阈值技术获取变化检测结果。真实的SAR数据集变化检测结果表明:该方法获取的差异图在噪声鲁棒性和对比度上均展现不错的优势,变化检测结果的kappa值在0.8以上,其可视化结果和评价指标均优于对比方法。

Considering the poor contrast between the changed and unchanged regions of the difference images of Synthetic Aperture Radar(SAR)images generated by the traditional algebraic operation algorithm and the weak noise robustness,we propose a neighborhood-level and superpixel-level difference images fusion method.Firstly,we obtain the superpixel difference images by superpixel segmentation of two-period images using superpixel combined with loga-rithmic operation;secondly,we fuse the similarity difference images and superpixel difference images in the neighbor-hood;finally,we obtain the change detection results using iterative thresholding technique.The change detection re-sults of the real SAR dataset show that the difference images obtained by this method exhibit good advantages in noise robustness and contrast,and the kappa values of the change detection results are above 0.8,and their visualization re-sults and evaluation indexes are better than those of the comparison method.