Research On Image Denoising Based On Partial Differential Equation And Wave Domain Methods

As an important carrier of information,the image plays an irreplaceable role in modern society of artificial intelligence.However,in the process of acquisition,transmission and storage,the image will be distorted in a degree by many kinds of noisy signal,which will affect the subsequent image processing.Therefore,in the area of image procedure,according to the cause of image quality's degradation,image denoising methods with the high performance is of great theoretical significance and application value in range of available device compute power and time.For some interior texture features and edge information of image,if algorithm of Partial Differential Equation is only diffused by gradient operator,the denoising method will not achieve ideal result.In transform domain,the denoising method has the character of sparse,however,which is easy to cause Gibbs effect.It is difficult for image de-noising method to how to effectively remove image noise while protecting the edge of the image and texture feature.In view of this,this dissertation mainly focuses on the denoising methods of the adaptive Gaussian Total Variation model and block matching in wave domain,starting from the angle of non-local.The main research contents are summarized as follows:(1)A hybrid weighted Wiener filter denosing model based on adaptive Gaussian Total Variation is proposed for protecting the information of image's edge and structure,and the weighted value ?,determined by SSIM,is introduced to balance the advantages of Wiener filtering method and an adaptive Gaussian Total Variation denoising method.The denoising effect of the model is verified by simulation.(2)In the wavelet domain,a three-dimensional block matching harmonic filtering model is established.The real signal of matching block group is represented in a sparse form by the correlation of three-dimensional transform,and then the purpose of de-noising is achieved by shrinking the threshold.Finally,the pre-estimated data of the image is obtained by inverse transform.The wavelet decomposition transform is used to extract the high-frequency part of the pre-estimated image to filter.To avoid edge ambiguity,Laplacian of Gaussian was used to construct a new operator into the diffusion model for filtering.Lastly,the final estimation of the original image is obtained by the wavelet reconstruction.(3)A new block matching denoising model based on Non-down Sampled Shearlet Transform is proposed.To avoid the occurrence of ill-posed problems,the scaling threshold is estimated by the histogram's statistical characteristics,and high-frequency sub-band is dealt by hard-threshold filter.Finally,the processed sub-band coefficients were inversely transformed and reconstructed to obtain the denoised image,and the feasibility of the new algorithm is verified by comparison experiments.