SAR Image Denoising Based On Bandelet Transform

Synthetic Aperture Radar (SAR) has characteristic of all-weather, day-and-night, high-resolution and strong-transmission, so it is widely used in civilian and military fields. SAR images reflect scattering characteristics of surface. Since SAR is a coherent-imaging system, SAR images are degraded by speckle, which considerably affects the detection of useful details, classification and the identification of certain objects. Therefore it is important to suppress the speckle in SAR images. Bandelet bases lead to optimal sparse representation for geometrically regular images and have potential power for the application of image denoising. An advanced algorithm based on Bandelet is presented, which is combined with Context Model and some good thresholds. The denoising result is satisfactory from evaluation index and visual measurement at the end. The principal tasks are as follows.This thesis, at first, introduces the principle of SAR imaging, and the characteristics of SAR images. Then based on these theories, it analyzes the cause of speckle in SAR images, and introduces some mathematical models for speckle reduction. Then it introduces several spatial-domain filters for specle reduction, and analyses experiments and tests results with real SAR images. Then we will focus on researching filtering algorithm based on Bandelet. Unlike most existing denoising algorithms, A key point of our approach is that, using contextual model to compute contextual values of Bandelet coefficients and then computing ideal thresholding according to these values. Experimental results using real SAR image demonstrate that the approach can remove the speckle noise efficiently and preserve edge of SAR image better.