The Research On Some Filtering Algorithms In The SAR Image Speckle

Synthetic Aperture Radar(SAR)is microwave imaging radar with high resolution.Because scattered signals produced by SAR system are processed coherently,SAR images are suffered from coherent interference inevitably.Speckle noise makes SAR images appearing a grainy appearance and impedes image calibration,inspection and interpretation,etc.The appearance of speckle seriously affects the quality of SAR images.So it is very important to suppress speckle noise for further applications of SAR images.There are two major categories of techniques for speckle reduction of SAR images.The first one is multi-look imaging process technology,which improves SAR image quality by averaging uncorrelated images from non-overlapping spectra,but this is at the expense of reduction of spatial resolution.The second one is image domain filtering technique and it suppresses speckle noise after the image is formed.It includes spatial domain algorithm based on local statistics,transform domain algorithm such as wavelet filter and diffusion filtering technique based on partial differential equations,such as anisotropic diffusion filtering algorithm.Spatial adaptive filter overcomes shortcomings of traditional non-adaptive spatial filter.It takes a sliding window on the image when reducing noise,then calculates pixel filter window weights and obtains local statistical characteristics of speckle images adaptively.It achieves good result and improves the quality of SAR image.And the algorithm is relatively simple,so it is commonly used in SAR speckle reduction.Based on the analysis of spatial adaptive filtering algorithm of SAR image,I propose a new speckle reduction method,main work is as follows.Firstly,Enhanced Lee filter is locally adaptive,it can better smooth SAR image based on fixed windows,but the retention effect of image edge information is not very well.Median filter is a nonlinear smoothing filter,and it has good ability to protect image details.This paper analyses multi-level median and it improves filtering algorithms especially.Based on the above study,this paper proposes a new algorithm,which using improved multi-level median and enhanced Lee filters,and experiments show that the new algorithm has better speckle reduction effect.Secondly,this paper analyzes wavelet transform,and introduces wavelet threshold speckle filtering algorithm on the basis of it.Threshold noise reduction algorithms are mainly soft and hard threshold method.It analyses several classical threshold estimation algorithms,and analyses Bayesian Shrink threshold function algorithms especially.Wavelet threshold de-noising algorithm may bring pseudo-Gibbs phenomenon.Circulated translation invariant algorithm could effectively remove pseudo-Gibbs phenomenon,which is its unique advantage.One new method is presented on the basis of Bayesian Shrink threshold and circulated translation invariant function.Experimental results show that the new algorithm improves the ability of the wavelet filtering,and overcomes the tilting and bending distortion characteristics of the in-phase axis after threshold reduced processing.