Abstract
Abstract. Block-matching and 3D filtering (BM3D) are used to reduce the multiplicative coherent speckle noise of Synthetic Aperture Radar (SAR) images, which may lead to the loss of image details. This paper proposes an improved similarity metric BM3D algorithm. Firstly, this method analyses the coherent speckle noise model, and applies a logarithmic transformation to make the BM3D algorithm suitable for multiplicative noise. Secondly, this method based on the calculation method of Euclidean distance weights for similar image blocks, and the Pearson correlation coefficient is introduced to improve the similarity metric. The accuracy of similar image block matching is improved, which is beneficial for removing image noise and maintaining image information. The experiments in this paper compared the results of this method with Frost filtering, Kuan filtering, wavelet soft thresholding and SAR-BM3D filtering algorithms. The results were compared and analysed by subjective vision and objective indicators. The experimental results show that compared with other filtering algorithms, the proposed algorithm has better ability to reduce speckle noise and preserve edge detail information for the image.
Highlights
Synthetic Aperture Radar (SAR) images are different from optical images dominated by additive Gaussian noise, which is dominated by multiplicative noise (Wang et al, 2017)
This paper uses the Pearson correlation coefficient to improve the Euclidean distance of the Block-matching and 3D filtering (BM3D) algorithm to find the most similar image blocks
The algorithm in this paper is compared with Frost filtering, Kuan filtering, wavelet soft threshold and SAR-BM3D algorithm in experiments
Summary
Synthetic Aperture Radar (SAR) images are different from optical images dominated by additive Gaussian noise, which is dominated by multiplicative noise (i.e. coherent speckle noise) (Wang et al, 2017). These methods can better maintain edge and detail information for the image, but the image will still be a lot of speckle noise Both types of filtering methods are calculated based on local models, which only consider local information of the image, and do not make full use of image structure information for denoising processing (Chen, Li, 2017). In 2017, Chierchia et al (Chierchia et al, 2017) used SAR-BM3D extension to process multi-temporal SAR data and proposed a multi-temporal synthetic aperture radar image removal speckle-noise algorithm based on block-matching and collaborative filtering This algorithm can suppress speckle noise and artefacts while retaining the image fine structure and region boundaries. Effectively remove coherent speckle noise and better retain the edge detail information of the image
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
