Research On Image Denoising Algorithm Based On Sparse Constraint

Image data is indispensable in people's daily communication.However,in the process of image transmission and reception,it is often interfered by noise due to the hardware equipment and other reasons,which will reduce the quality of image and affect the subsequent image processing and analysis tasks.Therefore,removing the image noise is very important.At present,how to protect the texture details of an image while removing noise is still an urgent problem to be solved.In recent years,the rise of sparse representation theory has made a great breakthrough in image denoising.Based on the sparse representation theory,we optimized and improved the sparse transform learning image denoising algorithm and the group sparse residual constraint image denoising algorithm.The innovative achievements are summarized as follows:Firstly,an image denoising algorithm based on graph laplacian regularization sparse transform learning is proposed.The proposed algorithm introduces the graph laplacian regularization,and embeds the geometric information of the image into the graph laplacian matrix,so as to protect the structure of the image from being destroyed while denoising.In addition,when measuring the similarity between image patches,the optimized sparse coding is combined to obtain better group low-rank estimation.Experiments show that the proposed algorithm not only has advantages in objective evaluation indicators,but also can better restore the structure of the image.Secondly,for the problem that the group sparse residual image denoising model cannot balance the denoising of the texture area and the smooth area,an image denoising algorithm based on the just noticeable difference weighted group sparse residual constraint is proposed.The texture area and smooth area of the image are measured by the just noticeable difference,and the fidelity items are weighted according to the value of the just noticeable difference.Since similar image patches often have similar sparse coding,and there is a certain correlation between similar image patches,a low-rank constraint is added for sparse coding to better utilize the structural information between similar patches and further improve the denoising effect of the image.Related experiments show that the proposed algorithm can better protect the texture details of the image and obtain better visual effects.Finally,a multi-image group sparsity residual constrained image denoising algorithm is proposed.Due to the randomness of noise,the similar image patches found in a single noise image are not accurate enough In order to solve this problem,the average image of multiple noise images is used to estimate the intensity of each image patch affected by noise,and each image patch is weighted according to the estimated noise intensity.The final distance between the two image patches is determined by the weighted sum of the Euclidean distance in multiple noise images.Considering that image patches with different similarities have inconsistent sparse residuals,the proposed algorithm reweights the sparse residuals according to the similarity.A large number of experiments show that this algorithm is not only better than single image denoising algorithm,but also better than multi-image denoising algorithm in objective evaluation indicators and visual effects.