Research On SAR Image Denoising Algorithm Based On Sparse Representation

Synthetic aperture radar(SAR)images are contaminated by multiplicative noise due to the defects of the inherent imaging mechanism during the imaging process,and the image noise hinders the subsequent processing such as target detection and identification.Therefore,the research on the denoising algorithm of SAR images is the focus of SAR field.Although classical image denoising algorithms are easy to implement,they have the disadvantages of unsatisfactory edge retention and incomplete noise suppression,and also make the original structure of the image lost to different degrees.In recent years,the development of sparse representation and image structure self-similarity theory has provided a new direction for image denoising.In this paper,we combine the above two methods and study new methods that can be applied to SAR image denoising based on sparse representation theory.(1)In this paper,a SAR image sparse denoising algorithm based on blind estimation and bilateral filtering is proposed.Firstly,bilateral filtering is used to obtain a preprocessed image with good edge-preserving characteristics,followed by blind estimation to obtain the full domain noise level of the image,which acts as the residual threshold in the sparse reconstruction process.Finally,sparse coding and dictionary learning algorithms are used to represent the image with as little atomic information as possible to achieve the purpose of image denoising.Using the noise level obtained by blind estimation as the sparse algorithm threshold,it is able to adaptively acquire the noise level of the image and smooth the image noise as much as possible.The experimental results show that the sparse reconstruction algorithm combined with blind estimation not only effectively removes the image noise and achieves a significant improvement in the equivalent view number,but also has good performance in peak signal-to-noise ratio and edge retention index,and effectively preserves the original image detail texture information.(2)The group sparse method combines the image nonlocal self-similarity(NSS),which has a better effect on the SAR image,a model with highly correlated image structure.Analyzing the image from the perspective of suppressing coherent spots,the image can be divided into homogeneous regions(flat areas of the image)and eterogeneous regions(detailed parts of the image edges),which have large gaps in structural information,and if the similar blocks are grouped directly the image blocks will be affected by each other.In this paper,we propose the Sub-region based group sparse representation algorithm(SR-GSR).Firstly,we use image probability statistical modeling to divide the image into two parts,and then divide the image into several selfsimilarity groups by block matching method,and use the similarity groups as the basic unit of denoising,which can generate adaptive intra-group dictionaries for denoising by natural image structure more effectively.It is experimentally demonstrated that the image obtained by sub-regional processing and then using group sparse reconstruction method can be found that the image quality is significantly improved after denoising,both visually and in terms of metrics.