Image Denoising Algorithm Based On Improved Non-local Means And Partial Differential Equation

With the progress of information age,image processing technology is more and more inseparable from our daily life.However,in the process of image acquisition,processing and transmission,the image will inevitably be affected by noise,resulting in the degradation of image quality.Noise removal is one of the most difficult problems in image analysis.Therefore,the algorithm that can effectively remove noise and protect the details of the original image has great research prospects.The image itself has structural redundancy and autocorrelation.The non-local mean algorithm uses the Euclidian distance of similar image blocks between pixels as the similarity measure to obtain a good de-noising effect,but the algorithm is difficult to meet the problem of preserving details and filtering noise at the same time,and the algorithm complexity is high.In this thesis,an improved structure tensor-based fast non-local mean algorithm FNLM is proposed.Image denoising algorithms based on partial differential equations are developing rapidly.The key is to construct appropriate equations.In this thesis,an improved hybrid model is proposed according to the advantages and disadvantages of YK model and PM model.The main research contents are as follows:1.An improved non-local mean algorithm is proposed.Firstly,the model proposed in this chapter replaces the weight function in the traditional non-local mean algorithm model,and improves the phenomenon of image blur after denoising.Secondly,by analyzing the eigenvalues and eigenvectors of the image structure tensor,the edge region and the smooth region of the image are divided,and then the parameters of different image regions are adjusted to enhance the denoising effect.Finally,the integrated graph acceleration algorithm is used to optimize the algorithm and improve the efficiency of the algorithm.The experimental results show that the algorithm proposed in this chapter can not only effectively remove the noise in the image,but also greatly improve the computational efficiency.2.A denoising algorithm of hybrid YK&PM model is proposed.Firstly,slice similarity module is introduced as edge detection operator in YK model,which makes up the flaw that Laplacian operator is too sensitive to noise in edge detection,and optimizes the performance of YK model to a certain extent.Secondly,due to the characteristics of anisotropic diffusion of PM model can solve the speckle effect well,and the characteristics of YK model can solve the common step effect in the second-order PDE model,a coupled denoising model of mixed YK&PM model is proposed,and a coupling coefficient is introduced to achieve the best denoising effect.The experimental results show that the algorithm proposed in this chapter has improved the ability of resisting noise,and has shown better results compared with other models.