SAR图像分割的改进参数核图割方法

Improved Parametric Kernel Graph Cut Method of SAR Image Segmentation

针对合成孔径雷达(SAR)图像的分割问题,提出一种改进的参数核图割方法。对参数核图割方法中的能量函数进行改进,在核空间中考虑分段常数模型,并实现目标函数的空间核化。SAR图像的分割通过能量函数的最小化实现,由不动点迭代估计区域参数,并由图割模型逐步最小化能量函数实现SAR图像的分割。为验证改进参数核图割方法的分割效果,对自然图像进行分割,结果表明,其分割精度达到83%,比参数核图割方法提高了11%。真实SAR图像的分割结果验证了该方法对SAR图像的分割结果优于参数核图割方法。

Based on the problem of Synthetic Aperture Radar（SAR） image segmentation, this paper proposes a new method. The method improves the energy function in the parametric kernel graph cuts, considering the piecewise constant model in the kernel space, and the kernelization of the objective function is realized in the feature space. SAR image segmentation is achieved ultimately by the energy function minimization. Energy minimization method is divided into two steps. Firstly, the region parameters are estimated by fixed point iterative algorithm. Secondly, the energy function is minimized by the graph cut method step by step and the image segmentation is realized. The natural images are segmented to verify the improved method of prarametric kernel graph cuts, and the segmentation accuracy reaches 83%, 11%higher than the parametric kernel graph cuts method. Real SAR image segmentation results demonstrate that the proposed method for SAR image segmentation result is better than parametric kernel graph cut method.