Total Variation Guided Filtering And Its Fast Algorithm For Image Denoising

Image is an important way of human communication.Because of the influence of various factors,different types of noise are introduced in the image,which leads to the decline of image quality.Filtering is a common method for image denoising.According to the image denoising space,the filtering methods are divided into two categories: one is the spatial domain filtering method represented by gauss filtering and median filtering,the other is the frequency domain filtering method represented by wavelet transform and fourier transform.Filtering method can suppress noise,but due to the inherent defects of the model,it may change the real pixels in the edge of the image,and reduce the structure preserve capability,affect the denoising result of the image.Then found that the PDE(Partial differential equation)denoising method can overcome the shortcomings of the traditional filtering method,which not only filter the noise effectively,but also keep the boundary of the image very well.Total variation using the gradient information as the regularization term,and processed by minimizing the model for image denoising.From the analysis of its model characteristics,it shows that it has strong edge preserve capability.Because of the gradient descent method is time consuming,which affect the calculation efficiency for total variation.Therefore,the split bregman algorithm is introduced.Not only the time is saved effectively,but also the result is better than the gradient descent method.To improve the edge protection capability of the bilateral filtering,a total variation guided bilateral filtering method is proposed.The result processed by the total variation model can maintain more structural information,provided more effective guidance information for the calculation of the range kernel function in bilateral filtering,so that the bilateral filtering can process the edge of the image accurately,and protecting more edge information.An iterative method of the above process is used to improve the robustness of the algorithm.With the improvement of the guided image,the performance of the bilateral filtering is also improved.To maintain a good denoising performance at different noise levels in guided filtering,an improved algorithm is proposed.Because the anisotropic total variation model used the first order norm of the gradient information as the regularization term,it can retain more edge information of the original image,and also can provide more structural information for the guided filtering.The experimental results indicate that the model breaks the limitation of the guided filtering,it can handle the condition of the higher noise level.