基于描述方法的SAR图像分割

SAR image segmentation based on description method

针对减少SAR图像分割中自由参数的问题,提出了基于最小描述长度的SAR图像分割方法。该方法经对数变换将SAR图像乘性噪声转换为加性噪声,对其建立描述模型,在描述长度最短意义上计算出重建图像,在假设SAR图像各区域实际地物后向散射特性对应的像素值恒定的前提下,该重建图像即为SAR图像的分割结果。该方法在分割的同时很好地抑制了SAR图像的相干斑噪声,保留了原始SAR图像的区域边界,并且不需要参数调节,整个分割过程自动完成,是一种非监督SAR图像分割方法。给出了该方法的具体实现步骤,实验结果验证了该方法的有效性。

To reduce the free parameters for synthetic aperture radar （SAR） images segmentation, this paper proposed aunsupervised segmentation method based on the MDL for SAR images. This method firstly established statistical models for log-arithm version of SAR images,then calculated the underlying images in the sense of minimum description length, which werethe segmentation images of the original SAR images under the supposition of the underlying piece-wise constant images. Thegood segmentation results were obtained with the region boundaries preserved well and the speckle noise was restrained perfect-ly. The SAR image segmentation can automatically be carried out without tuning any parameter, so the method was an unsuper-vised SAR images segmentation technique. It gave the detailed implementation process of the method, and verified the perform-ance of the method by both synthetic SAR images and real SAR images.