一种基于mean shift的Contourlet域SAR图像去噪方法

Contourlet变换是小波变换的新发展,具有良好的多尺度和多方向性。合成孔径雷达(SAR)图像Contourlet阈值去噪不考虑相邻像素在变换域的联系,将低于阈值的变换系数置零,会丢失图像中的细节信息。针对上述问题,文中提出一种新的去噪算法:首先,将SAR图像进行Contourlet分解;然后,利用具有良好间断点保留能力的mean shift算法处理子带系数。实验结果证实该算法能够在有效抑制相干斑噪声的同时,较好地保留图像中的细节信息。

A SAR Image Denoising Algorithm Based on Mean Shift in Contourlet Domain

Contourlet transform is a new extension of wavelet transform. It can provide a multi-scale and directional decomposition for images. In SAR image processing, the relationship between neighbor pixels is ignored usually by the Contourlet threshold method, and the coefficients less then the threshold value are set to be zeros derictly. Thus details in a SAR image are damaged. In this paper a new denoising algorithm is proposed to preserve more details. An image is decomposed into different subbands of frequency and orientation responses using the Contourlet transform. Then the mean shift, which can preserve discontinuity in the image, is used in the Contourlet domain. Experimental results show that details in the SAR images are preserved using the presented method while the speckle noise is reduced effectively.