Image Denoising Algorithm Based On The Nonlocal Block Matching

As one of the most intuitive ways of information dissemination,images play an irreplaceable role in most fields.However,due to the influence of the shooting and transmission equipment,the image will always be disturbed by noise,which will directly affect people's acquisition of the real image signal.And image denoising is the basis for further processing and understanding of the image.At present,these current denoising algorithms have achieved good results,but the preservation of image edges and detail information still faces enormous challenges.Under the nonlocal block matching framework,this paper researches on the searching similar blocks along the edge directions and the improved collaborative filtering based on the NCSR model.And this paper has obtained the following achievements:(1)In order to improve the sparse representation of a group of similar blocks of edge pixels,this paper proposes an image denoising algorithm based on the search of similar blocks in the edge direction.The algorithm first divides the image pixels into two categories:the edge pixel blocks are searched along the edge direction and the vertical direction,the smooth pixel blocks are searched along the horizontal direction and the vertical direction;the three-dimensional matrix is composed of the most similar matching blocks and the three-dimensional transformation is applied to the three-dimensional matrix,then,we can obtain a group of sparse coefficients of three-dimensional similar blocks;next,the hard thresholding filtering is used to suppress the noise;after a series of three-dimensional inverse transform,the image pixels obtain multiple estimates,and the weighted average of all the estimated values serves as the final denoising result of the pixel.The algorithm based on the edge directions searching similar blocks,which not only improves the similarity of similar blocks on the edge,but also accelerates the block matching speed,thus retaining more image details and improving the efficiency of the algorithm.(2)In order to further exploit the non-local similarity of one image and better describe the edge and texture,the algorithm improves the NCSR model to capture the image detail,and proposes a collaborative filtering denoising algorithm based on the NCSR model.Firstly,the smooth pixels and the edge pixels are respectively matching their similar blocks,and the three-dimensional groups are formed according to their Euclidean distances;the l1-norm of the NCSR model is used to process the three-dimensional group of smooth pixels;the l2-norm of the NCSR model is used to process the three-dimensional group of edge pixels;the multiple pixel estimates obtained from the NCSR filtering are aggregated to achieve the final denoising result.The algorithm improves the NCSR model and fully exploits the non-local characteristics of image blocks.At the same time,different filtering methods deal with the image edge pixel blocks and smooth pixel blocks,which better preserves the image edges and texture information and improves the performance of denoising.The experimental results,when compared with many start-of-the-art denoising algorithms,show that the proposed algorithms have a certain degree of improvement both in subjective visual effects and objective PSNR and SSIM values.