基于Gabor变换的雷达图像滤波及无监督分割

SAR Image Filtering and Unsupervised Texture Segmentation Based on Gabor Wavelet Transformation

提出了一种基于Gabor小波变换的多尺度、多方向的SAR图像去除斑点噪声及纹理分割算法．根据SAR图像的特点设计一组Gabor滤波器，对SAR图像进行二维Gabor变换，得到一组滤波后多分辨率、多方向的图像．通过对滤波后的图像分别进行非线性变换，再用非相干均值平滑滤出斑点噪声，并计算每个像素在选定窗口内的能量，以此检测出纹理特征，用均方误差聚类方法得到分割的图像．给出对SAR图像进行纹理分割的满意实验结果，对照试验表明，该方法优于空间灰度共现矩阵方法．

This paper presents a speckle reduction and texture segmentation algorithm for the SARimages with multiscale and multiorientation using Gabor wavelet transformation. A bank of Gabor filters is designed according to the specific features of the SAR image, and it is used to carry out 2D Gabor transformation of SAR images. A group of filtered images with multiresolution and multiorientation are obtained. After each (selected) filtered image has been subjected to nonlinear transformation, the speckles are then reduced with noncoherent median flatting, and the energy in specified window around each pixel is calculated. The texture features are characterized and the segmented imagesare produced based on a mean square-error clustering method. Experimental results of texture segmentation on SAR images are found to be satisfactory.