Research On Image Denoising Algorithm Based On Patches And Sparse Representation

Image denoising is not only an old subject problem,but also a hot research field in the moment.Many scholars are working on better algorithms to better repair noise images.With the advancement of modern science and technology,acquisition hardware is easy to handle images with excellent resolution and high shutter speed.But these factors also cause the image capture device to be easily corrupted by noise when capturing images.Effective image denoising technology can help camera manufacturers solve this problem.This also makes image denoising still a hot issue for continued research.This paper presents a new novel dictionary learning method for image denoising.The proposed idea is to incorporate the noise information into the design of a sparse coding algorithm,called improved sparse coding(Tikhonov-MOD-AK-SVD)that can effectively suppress the noise influence during the training.We utilize K-means method to group the noisy image patches.Each dictionary is trained by Tikhonov-MOD-AK-SVD in the corresponding class of images,which combine an overcomplete dictionary.Finally,the sparse coding technique is used to recover the image.Experimental results of image denoising on benchmark datasets show that the proposed method of dictionary learning is better than the traditional dictionary learning method.Secondly,we study the Nonlocally centralized sparse representation for image restoration.At present,the method of similarity between blocks based on Euclidean distance has some defects,ignoring the structural similarity between blocks.Based on this,we combine a measure of structural similarity to calculate the similarity measure between the blocks,and combine these two methods to calculate the sparse coefficient.The experimental results show that our algorithm achieves good results.Next,the non-convex Lp sparse optimization and the fixed point iterative theory are transferred to the image.We unify the group-based sparse coding i and the Schatten-p norm minimization problem by proving their mathematical equivalence.A fixed-point iteration scheme is developed for sparse optimization in lp space with p?(0;1] by using proximal operator and we a new solution to Schatten-p norm minimization problem is obtained.Finally,we study the fixed point iterative image denoising algorithm for non-convex Lp sparse optimization.we analyze the suitable setting of power p for each noise level.Through experiments,we set the appropriate p value under different noises to perform the algorithm.Experimental results demonstrate that for every given noise level,the proposed Spatially Adaptive Fixed Point Iteration(SAFPI)algorithm attains the best denoising performance on the value of Peak Signal-to-Noise Ratio(PSNR)and structure similarity(SSIM),being able to retain the image structure information,which outperforms many state-of-the-art denoising methods.When the noise is mixed noise,our algorithm still achieves good results.