A Sar Image Denoising Method Based on Nonlocal Constrained Sparse Representation

Abstract

The invention relates to an SAR image denoising method based on non-local restriction sparse representation. According to the SAR image denoising method based on non-local restriction sparse representation, dictionary learning is conducted with image blocks which are selected from a source image at random serving as training samples. When each image block is re-constructed, Gaussian kernel weights, based on similarity, are given to a plurality of image blocks which are located around the image block and are most similar to the image block, and only the representation coefficient of the current block is reserved to serve as a final result after similarity sparse representation is conducted. According to the SAR image denoising method based on non-local restriction sparse representation, the structure information in the source image can be effectively obtained through dictionary training, more image details are kept on the premise that effective denoising is achieved, and the better denoising effect is achieved.

Description

technical field [0001] The invention belongs to a digital image processing method, in particular to a SAR image denoising method based on non-local constraint sparse representation. Background technique [0002] Due to its special imaging principle, SAR images inevitably contain certain speckle noise, which affects the extraction and interpretation of image information. Therefore, it is very necessary to denoise SAR images before further processing. Traditional denoising methods mainly include spatial domain filtering, wavelet transform and other methods. However, these traditional methods often cause loss of detail information and incomplete denoising when denoising SAR images, and it is difficult to achieve good denoising effects. [0003] In recent years, sparse representation methods have been applied to image denoising. Compared with pixel-based filtering, image block-based sparse representation methods have natural advantages in preserving image structure information...

AI Technical Summary

Problems solved by technology

[0004] However, in SAR image denoising, due to the different imaging principles, it is often difficult to train a good dictionary from a noise-free natural image

However, the SAR image samples themselves contain noise, which makes the dictionary trained with these samples also contain noise.

This makes it difficult for existing sparse representation denoising methods to effectively remove noise while maintaining image details.

Images