基于纹理分类的极化SAR图像滤波方法

Parameter Estimation Method for PolSAR Filtering Based on Texture Classification

关于雷达图像优化,提高分辨率的问题,场景较为复杂的图像,固有噪声图像效果不够理想,对具有不同统计特性的像素点缺乏精确的区分。由于传统参数估计方法降噪效果不足,为解决上述问题,提出了一种基于纹理特征分类的参数估计方法。首先计算极化总功率图像的灰度共生矩阵,并提取纹理特征矢量,用K均值聚类的方法进行分类。然后根据分类结果,在滑动邻域窗内选取与中心像素同类别的像素用于参数估计。实验结果表明,改进的纹理分类的滤波方法具有更好的降噪效果,对于复杂场景的极化SAR图像表现了较大的优越性。

Polarimetric Whitening Filtering is a classical method for polarimetric SAR noise reduction, but the parameter estimation of the covariance matrix has always been a difficulty. The noise reduction effects of traditional methods, like the sliding neighborhood window, the Prewitt operator edge detection, and the structure inspection, are not good enough as they can not make a subtle distinction between the pixels with different statistical properties. To solve this problem, a new parameter estimation method based on texture classification has been proposed in this paper. Texture features were extracted from the span image, which then was used to calculate the gray-level co-occurrence matrix. Image pixels were then classified by K-mean clustering method. Parameters were calculated from the pixels of the same class in the sliding neighborhood window. Experiments demonstrate the effectiveness of this method. It shows much more advantage in polarimetric SAR images with complex scenes.