A Study Of Image Denoising Based On Nonlocal Total Variation Method

In present era,the development of information technology is particularly rapid.As an important part of information technology,image processing techniques have also been widely used.Using the variational method and partial differential equation to do the image processing is a new approac h that has just arisen in recent years,which has its unique advantages.However there are also some drawbacks to this approach,such as step effect and texture protection.Nonlocal variational method is the extension of image processing technology based on variational method and partial differential equation,which can solved step effect partly and performs well in texture protection.O n the one hand,the nonlocal method takes advantage of the similarity between image pieces to finish the image denoising.Due to the huge destruction caused by the multiplicative noise,the application of nonlocal method is greatly restricted.O n the other hand,the weighted Euclidean distance based on Gaussian kernel function,which is used in the traditional nonlocal method,is difficult to apply to the mult iplicative noise model.This will lead to the huge difference between the weight function and the real value.In addition,the large computation of nonlocal method is also a factor that restricts its development.In this paper,we will do some studies in terms of the above three aspects,which aims at the application of nonlocal variational method in image denoising.Firstly,for the contradiction between the denois ing characteristic of nonlocal method and the huge destruction caused by the mult iplicative noise,this paper presents a nonlocal variational model based on the gray level ind icator,which is used for image denoising on different gray levels.The model can remove the multiplicative well while avoiding step effect and protecting texture well.N umerical experiments also confirm the effect of this model.As the noise level goes up,the model has better abilities of texture protection than other local methods.Secondly,for inaccurate calculations of weight function caused by using Gaussian filter as the pretreatment of nonlocal method,this paper replaces Gaussian filter with AA model and AA model based on the gray level indicator,which obtain a good denoising effect.Especially when the noise level is not very high,AA model and AA model based on the gray level indicator have an incomparable advantage over Gaussian filter.Finally,the proposed models have been solved by the traditional steepest gradient descent scheme and the Split-Bregman iteration algorithm respective ly.The Split-Bregman iteration algorithm greatly improves the computational efficiency which can reduce the number of iterations from dozens or hundreds of times to less than ten times,while it does not affect the noise effect.This paper studies the image denoising based on nonlocal variational method.To remove the multiplicative noise,this paper presents a nonlocal variational model based on the gray level indicator.At the same time,the pretreatment method of nonlocal method is improved.In the end,the method of quick solution of the model is proposed.