Research Of De-speckling In SAR Images

Synthetic aperature radar (SAR) is an active microwave sensor.SAR can offer high resolution image and has the ability of clould penetration and weather independence. it has become one of the most powerful and efficient tools for investing and monitoring earth's environment.The use of SAR imagery has also been successfully extended to military applications, specifically for target detection and recognization. SAR is a coherent imaging system. So, speckle noise is a granular appearance in SAR image. Speckle noise makes it very difficult to visually and automatically interpret SAR data.Two types of approaches are traditionally used to suppress speckle in SAR image. The first is multilook processing, averaging of L multiple looks during the generation of the SAR images. The second approach, which is applied after the formation of the multilook SAR image, involves the algorithms based on estimation theory, the algorithms based on Partial Differential Equation (PDE) and the algorithms based on transform domain.This paper mainly studys the despeckling algorithms after the formation of SAR image. The despeckling filters based on estimation theory, PDE and SAR image heterogeneity measurement, are discussed in details in this thesis.1) We review the classic filters based on estimation theory in this paper. These speckle filters performance depends strongly on the speckle and scene models used as the basis for filter development. So, the parameter estimation of model distribution is the key technology here. The recently proposed "method-of-log cumulants" (MoLC), which is a parametric estimation methodology and relies on the Mellin transform, is introduced in our paper. Then, we propose a novel adaptive despeckling filter and derive a maximum a posteriori (MAP) estimator for the radar cross section (RCS), we model the RCS using the recently introduced General Gaussian (GG) density function. In our experiment, the new method was proved good performance in reducing speckle noise.2) Anisotropic diffusion is effectual in reducing additive noise. The theoretic foundation of applying anisotropic diffusion to Speckle suppression of SAR image is established through analyzing the statistical properties of the log-transformed speckle. It is based on the relationship between anisotropic diffusion and robust statistics and does well in preserving the detail of SAR image that the Biweight function is used as diffusion coefficient function in P-M. A new expression of gradient threshold is provided, which is a nonlinear decreasing function along time and protect edge information of image from blurring. The rule of stopping diffusion time, which is based on the max SNR, let the process to be efficient and exact. Experiments on SAR image show than the proposed method do well in speckle suppression and edge preserving, is a useful method.3) Conventional filters in the context of estimation theory, such as the Lee, Frost and Kuan, are usually based on the local coefficient of variation (CV). And another sort of speckle filter based on PDE, called speckle reducing anisotropic diffusion (SRAD) is also sensitive to CV. So, we conclude that the CV play a crucial role in the former two sorts of filter aimed at despeckling. The CV is a classical heterogeneity measure of SAR image. In our paper, the ratio of arithmetic to geometric mean (A/G, another heterogeneity measure) is used to replace CV and applied into the enhanced adaptive filters and SRAD. The novel kind of adaptive filters using for speckle, which are based on A/G, are proposed. Another despeckled algorithm, called SRAD based on A/G is also presented. Experiments validate the effectiveness of the proposed methods based on A/G. When facing an SAR image with intricate texture, the original filter based on CV do better performance the proposed filter based on A/G; when facing an image with characteristics of large edge, bright and simple texture, the A/G filter outperform the CV filter. So, the present filters based on A/G are supplementary to original filters based on CV.Finally, we make a summary of all proposed methods and describe the disadvantage of our algorithms. In addition, we point out the directions of further research on image registration.