基于Split Bregman算法的遥感图像分割分水岭技术

针对遥感图像中的道路、村落、农田、山脉等目标物分割时存在过分割和欠分割问题,采用分裂Bregman算法和分水岭方法相结合的方法,给出了适合近地遥感图像的一种新的标记分水岭方法。该方法利用多尺度插值小波算子逼近图像分割变分模型中的图像表示函数;采用Split Bregman迭代算法对变分模型进行快速求解;将Split Bregman分割结果作为标记,采用分水岭方法对遥感图像进行精确分割,保证了标记总数无冗余。试验结果表明,采用较小的尺度尺寸参数(J=3)时,可准确分割出来具有开环特性的村落图像区域,而采用较大的尺度尺寸参数(J=7)时,可精确分割农田中的纹理,但对村落区域则有一定的过分割现象出现。采用自适应多尺度参数时,可有效消除过分割和欠分割现象。

Watershed method on remote sensing image segmentation based on Split Bregman algorithm

Aiming at the problems of over segmentation and under segmentation problem in the remote sensing image of roads,villages,farmland,mountains,a method of Split Bregman algorithm and watershed method,a new method for surface marker based watershed remote sensing image,was used.This method uses multi-scale wavelet interpolation operator approximation image segmentation image model representation of the function;Split Bregman iterative algorithm is then used for fast solution of the variational mode;Finally,the method uses the segmentation results Split Bregman as marker,the watershed method for accurate segmentation of remote sensing image to guarantee the marking number without redundancy.The experimental results show that at the smaller scale parameter (J=3),a village image area open loop characteristics can be accurately segmented,while at the larger scale parameter (J=7),it can be accurate segmentation in the field of texture,but there are some of the over segmentation phenomenon appears on the village area.In conclusion,when the adaptive multi-scale parameters are used,the phenomenon of the over-segmentation and the under-segmentation can be effectively eliminated.