A Study Of SAR Image Denoising Based On Contourlet Transform

SAR is a coherent system which can produce high resolution remote sensing image and break through the limit of external condition effect. SAR is widely applied to military field and national economic, and its characteristics are as follows: all weather, all time,multi-band and multi-polarized working mode, all-variable side perspective, strong transmission.While SAR image segmentation is SAR image actuated processing foundation,but also one of the key technologies of automatic image understanding and interpretation..The traditional airspace and transform domain methods to deal with the noise of SAR image will lead to such problems as the detail information loss and noise removal is not completely in homogeneous area. It is difficult to effectively suppress speckle noise of SAR image, but the emergence of multi-scale geometric analysis domain statistical model provides a effective method for SAR image denoising processing.In this paper,I study the Contourlet transform method what is applied to SAR image denoising processing, proposed with the SAR image denoising algorithm which based on Contourlet transform and the kernel regression and achieving the better denoising result. The specific content and work as shown below.1. The study of SAR image denoising algorithm which based on Contourlet transform and the kernel regression. The method using Contourlet transform first Contourlet decomposed in SAR image, get the low frequency sub-band and high frequency sub-band images(coefficient), Then use kernel regression algorithm to remove the high frequency sub-band image noise and, increased-Lee filtering of low-frequency sub-band image denoising. Finally, use the Contourlet inverse transformation for low frequency and high frequency coefficients which after denoising, getting the denoised SAR image.2. Test an improved algorithm based on Contourlet transform named NSCT transform(non-sub-sampled Contourlet transform) and the kernel regression SAR image denoising algorithm. Because of Contourlet transform consists of Laplacian pyramid decomposition and direction filter banks, the transformation in the presence of samplingoperation, do not have translation invariance, denoising process will generate pseudo Gibbs phenomenon. So by using the non-sub-sampled contourlet transform(NSCT),combining the kernel regression algorithm, the denoising effect is better than the Contourlet + kernel regression algorithm.