Abstract:

Aiming at addressing the issue that the traditional method which employs gray level co-occurrence matrix as texture feature cannot effectively represent the high frequency information of image edge, this paper proposes a Synthetic Aperture Radar（SAR） image segmentation algorithm based on statistical characteristics of wavelet transform domain combined with multi-resolution analysis of wavelet transform. After wavelet transform, the statistical properties of the image obey the Generalized Gauss Distribution（GGD）. In this paper, two parameters of GGD, [α] and [β], are deduced by Maximum Likelihood（ML） estimation. A fast iterative solution of [β] is proposed by Newton-Raphson method. The texture features of SAR images are taken as [α] and [β], which are segmented by K- Means. The experimental results of typical SAR images show that this algorithm is effective.

Key words: Synthetic Aperture Radar（SAR） image, wavelet transform, Generalized Gauss Distribution（GGD）, Maximum Likelihood Estimation（MLE）, K- Means

摘要：

针对利用灰度共生矩阵作为纹理特征的传统方法不能够有效表征图像的边缘高频信息的问题，结合小波的多分辨率分析，提出了一种基于小波变换域统计特性的合成孔径雷达（SAR）图像分割算法。图像经过小波变换后，其统计特性服从广义高斯分布（GGD），利用最大似然（ML）估计，推导出GGD的两个参数[α]、[β]，提出了利用Newton-Raphson法对[β]进行快速迭代求解。并将[α]、[β]作为SAR图像的纹理特征，利用K-Means对其进行分割。通过对典型的SAR图像结果分析，表明了该算法的有效性。

关键词: 合成孔径雷达（SAR）图像, 小波变换, 广义高斯分布（GGD）, 最大似然估计（MLE）, K-Means