Research on Image Denoising Algorithm Based on Edge Enhancement Sparse Transform and Low Rank

Focused on the issue that detailed information of edge is easily lost in the process of image denoising, the image denoising algorithm based on edge enhancement sparse transform and low rank was proposed. Firstly, Canny algorithm was used to gain the image boundary to obtain the edge matrix, and secondly, the Non-subsampled contourlet transform (NSCT) was used to gain the multi-directional high-frequency sub-band map, and then the edge matrix was used to locate the edge position in the high-frequency sub-map. The edge sub-band coefficients were enlarged to enhance the edge to form the edge-enhanced noise image. Finally, the image was divided into image patches, then the local sparsity of image patches to sparse transform and non-local self-similarities of image patches to block match were used to achieve the purpose of image denoising. The experimental results show that, compared with K-Singular Value Decomposition (K-SVD), Group Low-Rank (GLR), Sparsifying Transform Learning and Low Rank (STROLLR), and DnCNN algorithms under different levels of noise, the proposed algorithm not only has a large improvement in the Peak Signal-to-Noise Ratio (PSNR), but also maintains clear edge details in visual effects while reducing noise.