Method Of Image Block Denoising Based On Adaptive Total Variation

The traditional L~2 norm variational image denoising model,also called harmonic model(TV2),is isotropic diffusion,namely the diffusion capacity is the same in all directions.Although the model can effectively remove the noise,it is easy to blur the edge of the image details.The classical L~1 norm total variational image denoising model,also known as ROF model(TV1),is anisotropic diffusion,which each point of the image is along the direction perpendicular to the image gradient,that is,spread along the image edge.Therefore,this model can effectively protect the edge of the image details.But it is easy for noise sensitive to misjudge the flat area of the noise as the edge,resulting in false edges and the phenomenon of staircase effect.Therefore,these two denoising models have their inherent advantages and disadvantages,using one of the models alone to denoise the image and can not achieve good results.Therefore,focusing on the analysis and research of the total variational image denoising model and aiming at the defect of the traditional total variational image denoising mode that is sensitive to noise and easy to blur,this paper proposes the method of image block denoising based on adaptive total variation,combined with the advantages of TV1 and TV2 these two variational denoising models.This method divides the image into flat region and edge region according to the local gray mean grads of the image,which can adaptively select the isotropic L~2 norm of the total variation image denoising model in the flat region or the anisotropic L~1 norm of the total variation image denoising mo del in the edged region according to the local gray mean grads.Moreover,we use the method of copying the neighboring pixels to fill the image border to solve the border processing problem of traditional total variation algorithm.In addition,this paper presents a method to adapt the optimal regularization parameter λ by studying the large number of experimental data and using the nonlinear least squares method to analyze the relationship between the optimal regularization parameter λ and the noise variance σ~2.In the end,a new non-reference image quality evaluation method,namely the average of local variance,which is used as an auxiliary index to evaluate the denoising ability,was introduced into this paper by Matlab simulated experiment.Experimental results showed that,when adding Gaussian noise of mean of 0 and variance of 0.01 to the blurred image,the method of image block denoising based on adaptive total variation can increase the Peak Signal-to-noise Ratio(PSNR)of the noise image by 11.82 dB.Compared with the traditional denoising algorithms,this method can not only preserve more texture details of the edge region,but also efficiently suppress the noise of the flat region.