Research On Image Denoising Algorithm Based On Non-local Self-similarity

In today’s era of artificial intelligence and big data,the transmission of information is no longer limited to the previous text mode,but is more based on high-speed communication methods such as voice calls,digital images,and video communications.Among them,images are the most direct and extensive way to obtain information in human daily life.However,in the process of acquisition or transmission of the imaging system,it is difficult to avoid noise interference,which can degrade the quality of images and thus make it difficult to carry out the work in several related fields in reality.Therefore,noise removal through image denoising techniques is a topic of great research significance.In recent years,the image denoising algorithm based on non-local self-similarity(NSS)theory has achieved a good denoising effect,which is a promising research direction in the field of image denoising.Based on the non-local self-similarity of images,this thesis proposes two improved image denoising algorithms combined with the weighted nuclear norm minimization(WNNM)algorithm.The main research work of this thesis is as follows :(1)A pixel-level Weighted Nuclear Norm Minimization(PWNNM)algorithm is proposed to solve the problem that the WNNM algorithm is susceptible to noise when calculating the similarity of image blocks while making greater use of the non-local prior information of images.Firstly,a pixel-level non-local prior is introduced to obtain a more accurate similarity block matrix when searching similar image blocks,and a noise level estimation method is proposed based on this matrix.Then the noise image is preprocessed based on Haar transform and Wiener filtering technology.Finally,the noise level is estimated on the preprocessed image,and the estimated noise variance is used to normalize the fidelity term of the WNNM algorithm to further denoise the residual noise and artifacts in the preprocessed image.The simulation results show that the proposed PWNNM algorithm can effectively remove the artifacts in the image,and the denoised image has a better subjective visual effect.(2)In order to better preserve the detailed features of the image,a WNNM algorithm combined with residual image completion is proposed based on the above pixel-level non-local prior by introducing the residual image theory.Firstly,the total variation(TV)model is used to denoise the noise image,and then the difference operation is carried out between the noise image and the initial denoising image.The noise residual image obtained by the difference is denoised by the improved wavelet threshold function,and the residual image after wavelet denoising is superimposed with the initial denoising image to obtain the residual completion image.Finally,the pixel level similar block matrix of the residual completion image is searched,and the WNNM algorithm based on residual noise level iteration is used for secondary denoising.The simulation results show that the proposed algorithm can effectively improve the objective index of the denoised image,can better maintain the image’s texture structure and is more effective in the high noise environment.